EVIDENTIALS AND INTERROGATIVES: A CASE STUDY FROM KOREAN

by Dong Sik Lim

A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (LINGUISTICS)

December 2010

Copyright 2010

Dong Sik Lim

Epigraph

“You attacked the reason… It’s a bad theology.” - G.K. Chesterton, ‘The Blue Cross’

“Ten false philosophies will fit the universe; ten false theories will fit Glengyle Castle. But we want the real explanation of the castle and the universe. But are there no exhibits?” -

!

G.K. Chesterton, ‘The Honour of Israel Gow’

""!

Dedication

To my parents

!

"""!

Acknowledgements

Looking back at my life in USC, after marking a ‘period’ to one stage of my life, I have to say: ‘I am nothing’, just as what Charles Schulz frequently said about himself (or this quote may be from Charlie Brown. I thought I saw this in David Michaelis’ recent biography on Schulz, but I could not find where it is. However, since Charlie Brown reflects the characteristics of the artist himself, it does not make much difference, I hope). The crucial difference between him and me is, of course, that I am not such a great person as Schultz. Unlike him, I am really nothing. Therefore, this thesis would also be just nothing, since it is written by a person who is nothing, and nothing can come from nothing (as you know, it is only God - or maybe FSM - who can ignore the first law of thermodynamics). Nevertheless, if this thesis can be ‘something’ at least to some people, it is not because of me, but definitely because of various people I have met during my life in USC as well as elsewhere, and because of their influences on me. This thesis is indebted from people I mentioned here as well as people I did not mention here. My deepest thanks go to my chair and advisor, Elena Guerzoni, from whom I was mostly influenced in my life in USC. Since the interview on phone before the admission to the PhD course in linguistics in USC, she has always shown her interests in my works. She gladly agreed to be my academic advisor, and after that, she read every draft of mine, including every term paper, no matter how poorly and roughly it was written. She showed no hesitance to take her time to have appointments with me, and suggested to me various new ideas to improve my papers. Sometimes it took much more time than expected to send her my papers due to various reasons, obviously sometimes including my laziness, and sometimes I made the same mistakes over and over again, but she always showed great patience and generosity, and corrected my mistakes over and over until I finally understand. She also emphasized how much it is important to present what

!

"#!

I know in an organized and understandable way, and always encouraged me when I began to lose confidence. Without her this dissertation could not appear. I am happy to have such a great advisor in my academic life. I really thank her for everything in USC. I also thank other committee members for their advices and guides. Andrew Simpson gladly agreed to be my committee member and gave useful comments on the final draft of this thesis for the defense. Jean-Roger Vergnaud joined my qualifying exam committee at the last moment, but he carefully read my drafts, gave many interesting examples, and asked many questions, which helped me think about the issues from different perspectives and improve the thesis. Stephen Finlay also gladly agreed to be the external member of my dissertation committee, even though he had never met me before. His comments greatly helped me to reconsider the relation between evidentiality and epistemic modality in Korean, and to make some crucial revisions on that part. Barry Schein was always interested in my work, and kept asking challenging questions and suggested to me new examples. Although he had to step down from the official committee for my defense to avoid potential problems in videoconference, he read all the drafts of my thesis, and gave me many useful comments and questions, which were greatly helpful while revising the thesis. I am also thankful to other faculty members in USC Linguistics. My thanks go to MariaLuisa Zubizarreta, who was always interested in my work, and her interest in Korean directed motion verbs and manner of motion constructions makes me keep working on argument structure and event structure of Korean predicates. Without Maria-Luisa I would forget and bury what I had done in the MA course in Korea. I am also thankful to Roumyana Pancheva, who took her time to discuss Korean scalar particles with me as my first screening committee member. Roumi also gave many useful comments when I wrote a paper about Korean deadjectival inchoatives, and her seminal paper on evidentiality guided me when I first began to work on this topic. I also thank Elaine Andersen, Sandra Disner, Toben Mintz, Barry Schein, and Rachel Walker for their

!

#!

encouragements and guides while I was TA-ing ‘Language and Mind’ for them. I also thank Sungbok Lee, who keeps me interested in computational linguistics. I also wish to thank professors in the department of linguistics in Seoul National University. Especially I thank Prof. Seungho Nam, who first taught me linguistics as instructor of ‘Introduction to Linguistics’ when I was a freshman, and was my MA thesis advisor. Because of his encouragements I could decide to study linguistics abroad. Even after I came to USC, he came to some conferences where I gave presentations and asked questions, and sometimes he gave me opportunities to present my works to students in SNU. I am really fortunate that I could learn linguistics and formal semantics from him, and that I have a great role model as a linguist and teacher like him. Prof. Chungmin Lee always showed great enthusiasm for linguistics, which I think I can never follow. He discussed with me any topics I brought to him, and gave useful comments no matter what the topic might be, which I really appreciate. I also thank other professors who gladly wrote recommendation letters for me when I applied to universities in US: Prof. Eun-Jung Yoo from Department of English Language and Literature, SNU, and Late Prof. Young-Jung Kim from Department of Philosophy, SNU. I would also like to thank other graduate students in USC. Students participating in ‘Syntax+’ and ‘SemPra’, including Janet Anderson, Priyanka Biswas, Katy McKinney-Bock, Mary Byram, Arunima Choudhury, Roger Liao, Sarah Ouwayda, and Barbara Tomaszewicz, always gladly listened to my presentations in various occasions and gave many useful comments and questions. I also thank my classmates in 2005, Wendy Cheng, Ed Holsinger, Nancy Louie, Phillip Potamites, Daylen Riggs, Rachel Sills, and Aaron Walker, for their friendships and helps in my study in USC. Special thanks go to Nancy Louie, who helped me a lot in TA-ing Ling 275lg in Spring 2008, and proofread several drafts of my other papers. I also thank Michael Shepherd for his proofreading the draft of this thesis, which was very complicated and poorly written at that time.

!

#"!

There are other people whom I should mention in the department of linguistics in USC, including Yin Bin, Candice Cheung, Xiao He, Stephanie Huang, Canan Ipek, David Li, Yi-Hsien Liu, Michal Temkin Martinez, Emi Mukai, Emily Nava, Megan Sensenbach, Erin Tavano, Laura Tejada, Hector Velasquez, Fuyun Wu, and Fang Ying. I also thank Korean colleagues in USC linguistics, Heeju Hwang, Heui-Joo Jeoung, Hyuna B. Kim, Lucy Kim, Eunjeong Oh, Eurom Ok, and Soyoung Park, for their helps with grammatical judgments and their friendships, and Joyce Perez for helping me to deal with all the administrative issues. Especially I thank Soyoung Park, who helped me to set up everything when I first arrived in LA. In addition to that, I am also indebted from audiences in SemPra (November 7, 2008, April 6, 2009, and April 12, 2010, USC), 18th Japanese/Korean Linguistics Conference (November 13-15, 2008, City University of New York), 38th Western Conference on Linguistics (November 21-23, 2008, UCDavis), and 2nd California Universities Semantics and Pragmatics (November 21, 2009, UCSC), where earlier versions of parts of this thesis were presented and I could receive some useful feedbacks. I also thank Gennaro Chierchia and Jon Gajewski for having an appointment with me and giving me useful feedbacks when they gave talks in USC, and Marcel den Dikken for attending my presentation in SNU and giving helpful comments and useful suggestions. It was always a great pleasure to chat with friends and seniors from the Department of Linguistics in Seoul National University about topics on linguistics as well as others. This group of colleagues includes Jinyoung Choi, Semoon Hoe, Hijo Kang, Midam Kim, Yoonshin Kim, Seongyeon Ko, Saera Kwak, Junkyu Lee, Seunghun Lee, Youngjoo Lee, Chulwoo Park, Jisung Sun, and Jiwon Yun. Especially, I thank Hijo, Jisung, Junkyu, Seongyeon & Jiwon, and Seunghun, for letting me stay in their places several times in various winter vacations and during several conferences. Special thanks go to Semoon, who always took his time to read rough drafts of almost every paper I have presented during my study in USC when I sent them to him, and gave me many helpful comments and questions. I can definitely say that outside USC, he is the

!

#""!

person who read my papers and drafts more than any other people. It must have been painful experience, I know - but he always did that gladly. I really appreciate it. I also have to mention some people in various Christian communities in Korea. I am afraid I have to painfully say that, I am very doubtful that I can be an ‘evangelical’ (in a strict sense, as historically and theologically defined by, e.g., Alister McGrath). This means that I may not share the same kind of ‘evangelical’ belief with them any more, even though I am still confessing Nicene Creed and Apostles’ Creed as my own confession of faith (which definitely means that I am still a Christian - even conservative). However, even though sometimes it becomes obvious that I take a position different from these people, they have always emotionally supported me and continued their friendships with me and it definitely helped me to survive the academic life in USC. Pastor Hyunchul Shin and other friends in Mapo Central Presbyterian Church in Korea always welcomed me when I visited Korea in vacations, even though I ‘fled’ from them in 2005, in some sense. I also thank three of old friends in Christian Association in Seoul National University in Spring 1999 - Mina Koo, Kwangsoo Mok, and Piho Park - for their friendships and emotional supports. These three are among few people who I think may understand why I changed. From Chihyun Park, Department of Sociology, SNU, I came to know that I am not the only one who is thinking about ‘something different’ from what we had believed and studied. I also thank Suna Choi, and Jonghye Woo & Sun Yu, for their continuous friendships with me. Sometimes, one simple phrase, ‘Thank you’, can tell much more than thousands of words. Last but not least, I would like to say that phrase to my mother and other family members in Korea, for their prayers, emotional supports, and their trust in me.

!

#"""!

Table of Contents

Epigraph

ii

Dedication

iii

Acknowledgements

iv

Table of Contents

ix

List of Figures

xii

Abstract

xiii

Chapter 1 Evidentiality, Contexts, and Questions: Overview 1.1. Evidentiality 1.1.1. Broad evidentiality 1.1.2. Narrow evidentiality 1.1.3. Speaker-dependency of evidentiality and its exception 1.1.4. Interim summary 1.2. Previous studies on the nature of evidential meaning 1.2.1. Evidentials as triggers of presupposition 1.2.2. Illocutionary meaning: Faller (2002) 1.2.3. Evidentials and tense morphemes 1.2.4. Interim summary 1.3. Semantics of indexicals 1.4. Conclusion and outline of the dissertation 1.5. Further discussion 1: interrogatives 1.6. Further discussion 2: an overview of evidential scales

1 2 2 4 13 14 15 16 18 25 30 31 37 38 41

Chapter 2 Evidentiality in Korean 2.1. Reportative evidentiality in Korean 2.1.1. The traditional view 2.1.2. Semantic diagnostics 2.1.2.1. Question and negation 2.1.2.2. First-person effects 2.1.3. Distributional diagnostics 2.1.3.1. Tense/aspect/modal morphemes 2.1.3.2. Subjects for quotative predicates 2.1.4. Interim summary 2.1.5. Is the pro-verb form also a reportative evidential? 2.1.6. Self-quotation: is -ay- a hearsay evidential? 2.1.7. Interim summary: non-final -ta- as reportative evidential

48 49 53 56 56 60 63 64 66 68 68 75 81

!

"$!

2.2. Direct evidentiality in Korean 2.3. Inferential evidentiality in Korean 2.4. Summary and conclusion

82 87 90

Chapter 3 Semantics of Korean Evidentiality 3.1. Korean evidential implication as presupposition 3.1.1. Question and negation 3.1.2. Truth/falsity of the assertion 3.1.3. Accommodation and correction 3.1.4. Interlude: the nature of the evidential presupposition 3.2. Korean evidentials and epistemic modals 3.2.1. Assent/dissent test 3.2.2. Embeddability 3.3. Summary and conclusion 3.4. Further discussion: speaker’s certainty and modal quantificational force

92 93 93 97 100 107 110 111 117 119 120

Chapter 4 Questions as Sets of Characters 4.1. Introducing the puzzle 4.1.1. Non-final -ta- as reportative evidential 4.1.2. Context-dependency of non-final -ta- and indexicality 4.1.3. The “evidential perspective shift” puzzle 4.1.4. Outline of the chapter 4.2. Kaplan’s theory of indexicals and Hamblin’s semantics of interrogatives 4.2.1. Kaplan (1989): pronouns as indexicals 4.2.2. Hamblin (1973): questions as sets of answers 4.3. A novel account for the evidential perspective shift in questions 4.3.1. Declaratives 4.3.2. Questions 4.4. A plausible (but incorrect) alternative: a pragmatic account 4.5. Review of previous accounts 4.5.1. Garrett (2001): questions as sets of assertions 4.5.2. Faller (2002): Cuzco Quechua evidential -si- in questions 4.6. Evidential -ta- vs. other perspective-anchored components 4.6.1. Expressives: optional shift 4.6.2. Utterance-modifying adverbs: optional shift 4.7. Korean direct evidential -te4.8. Conclusions 4.9. Further discussion: deriving a set of characters in Karttunen (1977)

130 130 130 133 134 135 137 137 142 146 148 150 153 156 157 159 166 167 168 169 181 181

Chapter 5 Remaining Issues and Further Implications 5.1. Sentential endings and evidential markers 5.1.1. Review of Kaplan (1989) 5.1.2. Back to evidentials in declaratives 5.1.3. Semantics of declarative endings 5.1.4. Semantics of question endings 5.1.5. Interim summary 5.2. The ‘intimacy’ reading of Korean reportative evidential -ta5.3. Evidentiality of sentential adverbials 5.4. Evidentials and epistemic modals

188 189 189 192 195 200 204 204 211 216

!

$!

5.4.1. Epistemic modals and narrow evidentiality 5.4.2. Cross-linguistic variations of evidentiality and epistemic modality 5.5. Reportative evidentiality in Korean noun complements 5.6. Summary and conclusions References

!

217 220 223 225 227

$"!

List of Figures Figure 1.1. Types of information sources in Willett (1988)

10

Figure 1.2. LF of (45a) Who left? in Hamblin (1973)

40

Figure 1.3. LF of (45a) Who left? in Karttunen (1977)

41

Figure 1.4. Types of information sources in Willett (1988) (repeated from above)

42

Figure 4.1. General schema of Kaplan’s (1989) two-step semantics

139

Figure 4.2. LF of Who left?

143

Figure 4.3. LF of (21) John-i na-lul po-ass-ta-nta

149

Figure 4.4. whether takes wide scope over -ta-

150

Figure 4.5. LF of (24) John-i na-lul po-ass-ta-ni?

151

Figure 4.6. LF of (73) John-i na-lul po-ass-ni? in Karttunen (1977)

183

Figure 4.7. LF of (76) John-i na-lul po-ass-ta-ni? in Karttunen (1977)

186

Figure 5.1. LF of (15) John-i na-lul po-ass-ta-nta with declarative ending -nta

198

Figure 5.2. LF of (23) John-i na-lul po-ass-ta-ni? with question ending -ni

202

!

$""!

Abstract

This thesis investigates the semantics and pragmatics of evidentiality in Korean and its interaction with the semantics of interrogatives. There is a rich and fruitful tradition of the study of evidentiality in typological studies of language and descriptive/functional linguistics, but works on evidentiality from the perspective of formal semantics only recently began to appear, and few of them address the question of how the semantics of evidentiality makes contributions to the semantics of interrogatives. My aims in this thesis are, therefore, to establish how evidentiality is grammatically encoded in Korean, and to investigate the semantic nature of evidential morphemes in Korean, which helps us to explain the semantic and pragmatic behavior of evidential markers in non-declarative sentences, such as interrogatives. By doing so, this thesis also shows the implications of Korean evidentials to cross-linguistic studies on evidentiality, and the semantics and pragmatics of interrogatives in general. This thesis is organized as follows. Chapter 1 is a general introduction to evidentiality. In this chapter I distinguish narrow evidentiality (evidentiality is expressed via a set of distinguished morphemes) from the broad evidentiality (evidentiality is expressed via linguistic objects whose main function is not directly related to evidentiality), and delimit the main issues that I will discuss in this thesis. Here I also outline two of previous approaches to evidentiality in formal semantics, that is, to analyze evidentials as epistemic modals triggering evidential presuppositions (Izvorski 1997, Matthewson et al. 2007, among others), and to analyze evidentials as illocutionary operators introducing evidentiality as illocutionary forces (Faller 2002, among others). Then I introduce the background theory I will adopt, that is, Kaplan’s (1989) theory of indexicals. Finally, I briefly outline the question I will mainly address in this thesis, that is, the evidential perspective shift in interrogatives.

!

$"""!

In Chapter 2 I argue that Korean is one of languages exhibiting the narrow evidentiality, and specifically, that Korean has (at least) two different evidential morphemes, the direct/perceptive te- and the indirect/reportative -ta-. Especially I focus on the indirect/reportative -ta- since, while there is a consensus that Korean -te- is a direct evidential morpheme (or, at least, that -te- is related to the direct evidentiality, either in the broad sense or in the narrow sense), it is still on debate whether Korean also has an indirect/reportative evidential marker, or there is only a phonological contraction of a indirect quotation which introduces the implication similar to that of reportative evidentials. Based on a set of semantic and morphological diagnostics, I argue that non-final -ta- in Korean is truly a reportative evidential marker in Korean. While doing so, I also argue against several previous works (Kim N.-K. 1990, 2001, Chung J.-Y. 2009, among others) which identify different morphemes as reportative evidential marker in Korean. Finally I will argue that Korean -te- should be analyzed as direct evidential marker, and even though it introduces implications which are similar to that of past-tense markers, it should not be analyzed as past tense marker (contra Chung K.-S. 2005, 2007). Chapter 3 concerns the semantic nature of Korean evidential markers. First I argue that evidential implications triggered by Korean -te- and -ta- are presuppositions, and not illocutionary forces, and in this sense, Korean evidentials behave in parallel with evidentials in languages like Turkish, Bulgarian, Norwegian (Izvorski 1997) and Stát’imcets (Matthewson et al. 2007), where evidentials markers are epistemic modals introducing evidential presuppositions. However, I also argue that, unlike Lee J.’s (2010) proposal, there is not enough evidence supporting the claim that Korean evidentials are epistemic modals, and that unlike languages where evidentials are epistemic modals, Korean evidentials do not have any semantic component of epistemic modals in their assertive meaning, but only trigger evidential presuppositions. This amounts to claiming that, typologically, Korean is a language which does not fit to any of the

!

$"#!

previous classifications: its evidentials are neither epistemic modals nor illocutionary operators, but merely presupposition triggers. Chapter 4 consists in the main proposal of this thesis. Here I introduce the puzzle of the evidential perspective shift in questions: evidentials are interpreted from speaker’s perspective in declaratives, but they are expected to be interpreted from addressee’s perspective in questions. Then I overview the background theories I am adopting here, that is, the semantics of indexicals proposed by Kaplan (1989) and the semantics of questions proposed by Hamblin (1973). Given this, I propose that Korean evidentials are functions from propositions to characters, introducing evidential presuppositions, and then I show that with this proposal we can derive the evidential perspective shift as a consequence of the interaction of the meaning of questions and the meaning of evidentials. Here I also show the advantages of this view over previous accounts developed in terms of pragmatics, and the differences between evidentials and other perspective-dependent components, like speaker-oriented adverbials and expressives. Furthermore, I discuss the implications of the analysis proposed for the semantics of interrogatives and of evidentiality in general. Chapter 5 is the conclusion of this thesis. Here I summarize the thesis, and address several remaining questions, including the role of sentential endings, the so-called intimacy reading / affectionate reading of Korean reportative evidential, the semantics of speaker-oriented adverbials in interrogatives, cross-linguistic variations of evidentials and epistemic modals, and reportative evidentiality in Korean noun complements.

!

$#!

Chapter 1 Evidentiality, Contexts, and Questions: Overview

The aims of this thesis are to provide a semantic analysis of evidentiality in Korean and to offer new insights into the semantics of interrogatives, evidentials, and characters. To achieve these goals, I investigate how evidentiality is grammatically encoded in Korean, the similarities and differences between Korean evidential markers and other related linguistic entities (such as epistemic modals and sentential endings), and how the semantics of Korean evidential markers interacts with the semantics of interrogatives. Although evidentiality has received much attention in descriptive linguistics and typological studies, formal semantic analyses of evidentiality have only recently begun to appear, and most do not address the questions raised in this dissertation. To answer these questions, this thesis provides a semantic analysis of Korean evidential markers that can account for their distribution in relation to other sentential markers and for their behavior in non-declarative sentences. Thus, the proposal made in this thesis carries important implications for the semantics and pragmatics of questions, modality, and evidentiality in a broad sense. This introductory chapter is organized as follows. In Section 1, I introduce the notions of broad and narrow evidentiality, and I discuss the main issues in the semantics and pragmatics of the evidentiality that this thesis is concerned with. In Section 2, I outline previous formal approaches to evidentiality, focusing in particular on how each theory analyzes the semantic nature of narrow evidentiality encoded in the evidential markers of a language. In Section 3, I briefly introduce Kaplan’s (1989) theory, which I adopt as a background assumption to address the questions raised in this thesis. I also illustrate the main focus of this thesis, namely, the interaction between the semantics of evidentials and the semantics of interrogatives. In Section 4, I summarize this chapter and outline the rest of this thesis. Finally, I discuss a couple of ! 1

secondary issues: Hamblin’s (1973) semantics of questions, which is not directly related to the notion of evidentiality but is closely related to the main proposal of this dissertation, and the notion of the evidential scale, which is one of the most widely discussed topics in the evidentiality literature but not a primary focus of this thesis.

1.1. Evidentiality

In this section I provide an overview of the notion of evidentiality. Specifically, I distinguish broad and narrow evidentiality and discuss what these notions have in common. I also posit that Korean has narrow evidentiality, a claim for which I provide evidence in Chapter 2.

1.1.1. Broad evidentiality

Broad evidentiality refers to any indication regarding the information that grounds a speaker’s assertions in terms of their source and degree of certainty. When a speaker makes an assertion, it is based on types of certain information sources, which include the following:

(1)

I. VISUAL: acquired through seeing. II. NON-VISUAL SENSORY: acquired through hearing (typically extended to smell, taste, and sometimes touch). III. INFERENCE: based on visible or tangible evidence or results. IV. ASSUMPTION: based on evidence other than visible results, including logical reasoning, assumptions, or general knowledge. V. HEARSAY: reported but with no reference to who reported it. ! 2

VI. QUOTATIVE: reported and including an overt reference to the source. (Aikhenvald 2004: 63)1

In most cases, the speaker does not need to specify the type of information source the assertion is based on, but natural languages offer many grammatical devices to do so if the speaker chooses to. The following examples show how a speaker can indicate the source of information in English.

(2)

a. I saw that John beat Mary. b. Bill said, “John beat Mary.” c. Apparently John beat Mary. d. John must have beaten Mary.

In (2a), the matrix clause I saw that indicates that the speaker saw that John beat Mary.2 In terms of (1), this means that the speaker has visual evidence of the validity of the embedded clause. In (2b), the speaker claims to have reportative evidence from Bill. In (2c), the adverb apparently suggests that the speaker based the assertion on some sort of inference. For example, the speaker may have concluded that John beat Mary based on the fact that John hates Mary or that Mary has bruises on her face. Initially, it seems strange that a speaker’s inference can serve as ‘evidence’ of

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 1

Aikhenvald (2004) summarized (1) as types of evidence that are grammaticalized in languages exhibiting narrow evidentiality (a notion that I will discuss in the next subsection), but these classifications can also be thought of as types of evidence on which assertions can be based in general. 2 (2a) is actually similar to examples in Simons (2007), where she discusses cases where verbs such as discover, believe, and suggest do not introduce any factive presuppositions, but rather serve some discourse functions, specifically indicating evidentiality. For a syntactic analysis of the evidential function of the main clause, see Rooryck (2001a, 2001b).

! 3

an assertion, but this fits the definition of broad evidentiality. (2d) may also express inferential evidentiality; based on certain indirect evidence, the speaker infers that John beat Mary.3 To sum up, we take from the examples in (2) that broad evidentiality can be conveyed grammatically in a wide variety of ways, including embedding predicates, epistemic modals, and speaker-oriented adverbials, just to name a few. In principle, every natural language can express evidentiality in these ways.

1.1.2. Narrow evidentiality

My focus in this thesis is on a more interesting cross-linguistic phenomenon: narrow evidentiality, which can be defined as “a linguistic category whose primary meaning is the source of information.” (Aikhenvald 2004: 3). As we saw in the previous subsection, any language can express broad evidentiality, but some languages develop specific linguistic categories to express evidentiality. In other words, their lexicons include specialized morphemes whose defining characteristics are listed in (3).

(3)

a. Evidentials show the type of justification for a factual claim that is available to the person making that claim. b. Evidentials are not the main predication of the clause, but a specification added to a factual claim about something else. c. Evidentials have the indication of evidence as their primary meaning, not only as a pragmatic inference.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 3

(2d) is an example of the ‘evidential’ use of the epistemic modal; see von Fintel and Gillies (2007, 2010) for further discussion.

! 4

d. Morphologically, evidentials can be inflections, clitics, or other free syntactic elements, but not compounds or derivational forms. (Faller 2002: 5; Originally from Anderson 1986: 274-275)

One language with evidentials that satisfy these criteria is St’át’imcets (also known as Lillooet, a variety of Northern Interior Salish). St’át’imcets has the three evidential morphemes shown in (4), corresponding to three types of evidence: reportative, inferential, and (inference from) perceived evidence. Examples of their usage are given in (5).4

(4)

Evidentials in St’át’imcets Evidential

Gloss

ku7

reportative

k’a

inferential

-an’

(inference from) perceived evidence (Matthewson et al. 2007: 202)

(5)

a. wa7 be

ku7

ku

REPORT DET

sts’éts’quaz’

l-ta

stswáw’cw-a.

trout

in-DET creek-EXIS

“[Reportedly] there are trout in the creek.” b. plan

k’a

tu7

already INFER then

wa7

tsu7c

na

máq7-a

IMPF

melt(INCH)

DET

snow-EXIS

“The snow must have melted already.” !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 4

As pointed out by many authors working on evidentiality (Faller 2002, Aikhenvald 2004, among others), evidentials are extremely difficult to translate. We must keep in mind that translations usually do not reflect the actual linguistic nature of evidentials in the language under investigation. For example, in (5a), even though the evidential meaning is translated as the sentential adverb reportedly, it does not mean that the reportative evidential ku7 in St’át’imcets is a sentential adverb.

! 5

c. pel’p-s-ácw-an’

nelh

lost-CAUS-2SG.CONJ-PERC.EVID DET.PL

nelklíh-sw-a key-2SG.POSS-EXIS

“It looks like you’ve lost your keys.” (from Matthewson et al. 2007: 202, 203)

In (5a), the evidential morpheme ku7 indicates that the speaker has heard someone assert the ‘prejacent’.5 Similarly, in (5b), the speaker’s assertion is based on inference, and in (5c), the speaker has direct perceptual evidence (e.g., visual) concerning the prejacent. Since the primary function of these morphemes is to indicate what type of evidence the speaker’s assertion is based on, they can be analyzed as evidential morphemes in St’át’imcets. Now, let us see in more detail how the criteria in (3) distinguish narrow evidentiality from broad evidentiality. First, consider (3a):

(3) a. Evidentials show the type of justification for a factual claim that is available to the person making that claim.

(3a) is related to the definition of evidentiality in general: evidentials, like other expressions exhibiting broad evidentiality, express the source of information that justifies the speaker’s assertion. This is not crucial to distinguish between broad and narrow evidentiality. (3b) is the most important characteristic of narrow evidentiality:

(3) b. Evidentials are not the main predication of the clause, but a specification added to a factual claim about something else. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 5

Throughout this thesis, I extend the notion of prejacent to indicate the propositional content of the sentence abstracted from the context of evidentials. I use this term in a purely descriptive way.

! 6

In terms of type-driven semantics, (3b) can be rephrased as follows: evidential meaning is not directly asserted, so it is not truth-conditional. This criterion separates the English case in (2a), repeated as (6a) below, in which the embedding clause expresses broad evidentiality, from the typical example of narrow evidentiality in St’át’imcets given in (5c), repeated as (6b) below:

(6) a. I saw that John beat Mary. b. pel’p-s-ácw-an’

nelh

lost-CAUS-2SG.CONJ-PERC.EVID DET.PL

nelklíh-sw-a key-2SG.POSS-EXIS

“It looks like you’ve lost your keys.”

In (6a), as we saw above, the embedding verb combined with the first person subject expresses evidential information. However, even though the embedding clause I saw that expresses broad evidentiality (indicating that the claim that John beat Mary is based on the speaker’s direct visual perception), it is not by itself an evidential because it is still a part of the assertion. If the predicate combines with another subject, as in (7a), we do not have broad evidentiality any more, and the evidential meaning can be negated, as in (7b). Both show that the evidentiality in (6a) is only the broad one and is still truth-conditional.

(7) a. Tom saw that John beat Mary. b. I did not see that John beat Mary.

However, as illustrated in Matthewson et al. (2007), the evidential marker an’ in (6b) does not contribute to truth-conditions, but only indicates the source of information. For example, unlike (7b), the evidentiality in (6b) cannot be negated, as shown in (8): ! 7

(8)

cw7áoz-as-an’

kw s-nilh-ts

NEG-3CONJ-PERC.EVID DET NOM-FOC-3POSS

s-Sylvia

ku

xílh-tal’i

NOM-Sylvia

DET

do(CAUS)-TOP

= “It is necessarily not Sylvia who did it.” [based on indirect perceived evidence] NOT: “It is not necessarily Sylvia who did it.” [based on indirect perceived evidence] NOT: “I don’t have indirect perceived evidence that it was necessarily Sylvia who did it.” (Matthewson et al. 2007: 219, ex. 42)

In (8), only the proposition that Sylvia did it can be negated; the evidential meaning that the assertion is based on indirect perceived evidence cannot be negated (putting aside, for the moment, the second possible scope of negation). This example shows that the meaning introduced by the evidential marker in narrow evidentiality is not truth-conditional, and the evidential marker does not contribute to the truth-conditional meaning. Before we discuss (3c), however, note that there are some exceptions. For example, in many languages, epistemic modals introduce evidentiality. According to Matthewson et al. (2007), the evidentials in Lillooet illustrated in (4) and (5) are epistemic modals, introducing evidentiality as presuppositions. However, as Matthewson et al. (2007) shows, even in these cases, the evidentiality is not-truth-conditional. Even though the epistemic modals are truth-conditional in the standard semantics of modals (e.g., Kratzer 1977, 1991), their evidential ‘implications’6 are non-truth-conditional.7 The implication introduced by speaker-oriented adverbs, such as apparently in (2c), is, in this sense, very close to evidential implication, leading us to ask whether !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 6

Throughout this thesis, the term implication is used to avoid making any theoretical commitment to the nature of the evidential meaning, i.e., whether it is truth-conditional, presuppositional, implicational, or illocutionary. 7 One might ask whether English epistemic modals could also be treated as evidentials. As mentioned above, some, including von Fintel and Gillies (2007), have pointed out that English epistemic modals introduce evidential implications. The problem is that English modals seem not to give enough information about the type of information source, which is the defining characteristic of narrow evidentiality. However, as we will see later, there are languages where only two types of information sources - direct and indirect - are distinguished, and English may be this type of language. For the moment, I leave this question open.

! 8

this kind of sentential adverb can also be treated as an evidential, but I leave this question open in this thesis.8 Let us turn now to (3c):

(3) c. Evidentials have the indication of evidence as their primary meaning, not only as a pragmatic inference.

(3c) guarantees that evidential implications are introduced by specialized morphemes and not by other lexical items whose primary meaning is unrelated to the notion of evidentiality.9 In (6a), for example, I saw John beat Mary, the primary meaning of the main predicate I saw is simply that the speaker saw something. As in (7a), when the predicate is used with another subject, its primary meaning appears overtly. Finally, consider (3d):

(3) d. Morphologically, evidentials can be inflections, clitics, or other free syntactic elements, but not compounds or derivational forms.

(3d) requires that narrow evidentiality be encoded in independent morphemes and not obtained by combining lexical items. For example, consider the embedding predicate see again, it can express evidentiality only when combined with the first-person subject. Summing up the above discussion, in the spirit of Anderson (1986), I use the following criteria to identify evidentials in a given language:

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 8

I will return to this issue in Chapter 5, but for the non-truth-conditional semantics of sentential adverbs, see, e.g., Potts (2005). 9 Perhaps this is why the only exceptions to (3b) are epistemic modals with evidential implications, because the primary meaning of epistemic modals concerns the speaker’s state of knowledge, which is closely related to evidentiality.

! 9

(9) a. An evidential morpheme encodes the information source the speaker’s utterance is based on.10 b. An evidential morpheme is morphologically independent and atomic. c. The information about the type of evidence expressed by evidentials is not truthconditional.11

The next question is what type of evidence can be grammaticalized as narrow evidentiality. There have been many proposals regarding this question, one of which is Willett’s (1989), based on his investigation of 38 languages:

Figure 1.1. Types of information sources in Willett (1988) Types of Evidence q p Direct | Attested e|i visual auditory sensory

Indirect q p Inferred Reportative ei e | i result-based reasoning-based second-hand third-hand folklore (from Lee J. 2010)

Willett (1988) proposes this hierarchy to cover all possible inventories of evidentials in natural language, claiming that all cross-linguistic variation in evidentials can be accounted for in terms of which category in this hierarchy is grammaticalized and how the hierarchy is divided up. For !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 10

As we will see in detail in Chapter 4, however, this requirement is different when the evidential morpheme is used in questions. When used in questions, the evidential morpheme (at least in Korean) indicates what type of evidence the addressee’s answer is based on. 11 One might ask whether the embeddability is also included in the criteria of narrow evidentiality or not. It has been reported that many evidentials cannot be embedded (e.g., Cuzco Quechua evidentials in Faller 2002), but other evidentials, such as St'át’imcets evidentials, can be embedded (Matthewson et al. 2007). Therefore I do not include embeddability as a criterion for narrow evidentiality. For the embeddability of Korean evidentials, I will postpone the relevant discussion until Chapter 3.

! 10

example, Cherokee only distinguishes direct and indirect evidence, with !"i in (10) indicating that the speaker’s assertion is based on direct evidence (the speaker saw it, heard it, etc.) and e"i in (11) indicating that the assertion is based on indirect evidence (evidence from what the speaker heard or an inference from what the speaker saw or heard, etc.).12

(10) a. wesa cat

u-tlis-!"i it-run-FIRST.PAST

‘A cat ran.’ Scenario: I saw the cat running (Aikhenvald 2004: 26, ex. 2.3.) b. un-atiyohl-!"i they-argue-FIRST.PAST ‘They argued.’ Scenario: I heard them arguing. (Aikhenvald 2004: 26, ex. 2.4.)

(11) a. u-wonis-e"i he-speak-NON.FIRST.PAST ‘He spoke.’ Scenario: Someone told me. (Aikhenvald 2004: 27, ex. 2.7.) !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 12

Unlike Aikhenvald (2004), which does not distinguish in the translation the situation in which the evidential in the example can be used felicitously, in the translation, we only indicate the truth-conditional meaning, and under ‘scenario’ we indicate the situation where the given example is used felicitously.

! 11

b. u-gahnan-e"i it-rain-NON.FIRST.PAST ‘It rained.’ Scenario: I woke up, looked out, and saw puddles of water (Aikhenvald 2004: 27, ex. 2.8.)

Different languages may have more fine-grained systems of (narrow) evidentiality. For example, Wintu (a Native American language spoken primarily in the Northern California) distinguishes five types of evidence sources: visual (be.), non-visual sensory (nthe.), inferred (re.), assumed ("el.), and reported (ke.). Examples are given in (12).

(12) a. k’upa-be. ‘he is chopping wood’

=> VISUAL

Scenario: I see or have seen him. b. K’upa-nthe. ‘he is chopping wood’

=> NON-VISUAL SENSORY

Scenario: I hear him or a chip flies off and hits me. c. K’upa-re. ‘he is chopping wood’

=> INFERRED

Scenario: I go to his cabin and find him absent and his axe gone. d. K’upa-"el. ‘he is chopping wood’

=> ASSUMED (EXPERIENTIAL)

Scenario: I know that he has a job chopping wood every day at this hour, that he is a dependable employee, and perhaps that he is not in his cabin. e. k’upa-ke. ‘he is chopping wood’

=> REPORTED

Scenario: I know from hearsay. (Aikhenvald 2004: 27, ex. 2.96)

! 12

One of the aims of this dissertation is to show that a set of morphemes in Korean satisfies the criteria proposed in (3) and (9) and, thus, that Korean exhibits narrow evidentiality and falls within the typology suggested by Willett (1988). Specifically, in Chapter 2, I argue that Korean distinguishes direct and reportative evidentiality through the use of the morpheme -te- for direct evidentiality and -ta- for reportative evidentiality and possibly also inferential evidentiality via other morphemes.

1.1.3. Speaker-dependency of evidentiality and its exception

There is cross-linguistic variation in the semantic objects that evidential markers express and sometimes variation even within a given language. However, in all languages encoding evidentiality in the broad or narrow sense, it has one common feature: evidentiality is always speaker-oriented. This feature plays a fundamental role in my proposal about the semantics of evidentiality in Korean. To make this more concrete, consider example (10a), repeated as (13) for convenience, in which the evidential implication is that the speaker saw a cat run (or the speaker has direct evidence that a cat ran); that is, the direct evidence is the speaker’s and no one else’s.

(13)

wesa

u-tlis-!"i

cat

it-run-FIRST.PAST

‘A cat ran.’ Scenario: I saw the cat running (Aikhenvald 2004: 26, ex. 2.3.)

! 13

This holds not only in narrow evidentiality, but also in broad evidentiality.13 Returning to the previous examples, repeated as (14), the fact that we have the evidential implication in (14a) but not in (14b) can be accounted for in terms of the speaker-dependency (or speaker-anchoring) of evidentiality in general. Thus, unlike in (14a), in (14b), the person who saw John beat Mary is not the speaker. Hence the absence of the evidential implication.

(14) a. I saw that John beat Mary. b. Tom saw that John beat Mary.

The speaker-dependency of evidentiality is reminiscent of the speaker-dependency of indexicals, such as the pronoun I or the adverbs here and now. These expressions are also speaker-dependent, in the sense that their interpretations vary depending on who speaks them. However, in Section 1.3, I also show that there is a crucial difference between evidentials and other indexicals, namely that the speaker-dependency of evidentials changes when they are used in interrogatives, whereas that of indexicals does not vary. Illustrating why such a difference exists is another aim of this thesis. I return to this issue in Section 1.3 and then illustrate it in detail and analyze it in Chapter 4, where the main proposal of this thesis is presented.

1.1.4. Interim summary

In this section, I briefly outlined the notion of evidentiality, distinguishing broad and narrow types. I also introduced a set of detailed criteria to distinguish between the two. Finally, I pointed out that the central characteristic of both broad and narrow evidentiality is speakerdependency. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 13

Higginbotham (2009) also addresses the speaker-dependency of evidentiality under the term first-person authority.

! 14

1.2. Previous studies on the nature of evidential meaning14

In the previous section, I pointed out that one of the main characteristics of narrow evidentiality is that it does not contribute a truth-conditional component. There are, however, several phenomena related to meaning in natural language that cannot be reduced to truthconditions, including conversational implicature (e.g., Grice 1975), conventional implicature (e.g., Grice 1975, Potts 2005), presupposition (e.g., Soames 1989), and illocutionary force (e.g., Searle 1969, Searle and Vanderveken 1985). Therefore, when claiming that a language, such as Korean, exhibits narrow evidentiality, we need to be specific about the type of non-truth-conditional contributions evidentials in that language make. The literature on this question divides into two camps. One treats narrow evidentiality as presupposition (e.g., Izvorski 1997, Chung K.-S. 2007, Matthewson et al. 2007, Rullmann et al. 2008), and the other views it as encoded in illocutionary operators (e.g., Faller 2002, 2006). There is also an intermediate position, where evidentiality is encoded as epistemic modals in some morphemes and as illocutionary operators in others (e.g., Garrett 2001). In this introduction, I focus on the first two options. Finally, in addition to these three camps, a new approach recently emerged in which some niches with the landscape of evidentiality have been characterized in terms of a special-kind of tense semantics (e.g., Faller 2004, Chung K.-S. 2007). In this section, I review these three possibilities in turn.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 14

Cf. Matthewson et al. (2007, esp. Section 3) for an overview.

! 15

1.2.1. Evidentials as triggers of presupposition

The camp that analyzes narrow evidentiality as presupposition claims that there is a truthconditional component in the meaning of evidentials which coincides with that of epistemic modals (e.g., Izvorski 1997, Matthewson et al. 2007).15 In addition to this, the evidential presupposition is treated, in standard terms, as a felicity condition, which is technically imported as a definedness condition (Heim and Kratzer 1998). In other words, the propositional content of the prejacent can be felicitously uttered if and only if the felicity condition is satisfied. As an example of proposals in this camp, let me illustrate Izvorski’s (1997). She begins by observing that in languages from various language groups, including Turkish, Norwegian, and Bulgarian, the present perfect morpheme or a morpheme that originates from it may express indirect evidentiality (i.e., that the speaker’s assertion is based on indirect evidence, either reportative or inferential):16

(15) a. Gel-mi!-im.

(Turkish)

come-PERF-1SG b. Az I c. Jeg I

sâm

do"al.

be-1SG.PRES

come-P.PART

har

kommet.

(Bulgarian)

(Norwegian)

have-1SG.PRES come-P.PART

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 15

Some researchers from functional perspectives claim that epistemic modals do not contribute to propositional content (and therefore are not truth-conditional). Given this approach to epistemic modals, one can see that treating evidentials as epistemic modals does not necessarily converge into treating evidential implications as presuppositions, or vice versa. Epistemic modals may introduce only evidential implications on a level different from that of truth-conditions, such as illocutionary meaning, or evidentials may introduce only the presupposition without affecting the propositional content of the prejacent. However, since most works treating evidentials as epistemic modals also treat evidential implications as presuppositions, I do not strictly distinguish these two approaches in this introduction. For discussion of the truthconditional contributions of epistemic modals, see, e.g., Papafragou (2006) and Portner (2009). 16 For simplicity, I do not discuss the temporal and aspectual semantics of PE. I will return to her analysis in Chapter 3, but see Izvorski (1997) for a detailed analysis.

! 16

‘I have come.’

(present perfect)

and/or

‘I apparently come.’

(Perfect of Evidentiality; PE) (Izvorski 1997: 222, ex. 1)

To account for this, Izvorski (1997) proposes that PE is a universal epistemic modal corresponding to English must, which introduces an evidential presupposition. Consider (16) and (17):

(16) a. John must have drunk all the wine. b. But I have no evidence for that. c. (#But I have no reason for believing that.) (Izvorski 1997: 228, ex. 12)

(17) a. Ivan Ivan

izpil

vsi#koto

vino

v#era.

drunk-PE

all-the

wine

yesterday

‘Ivan apparently drank all the wine yesterday.’ b. #But I have no evidence for that. (Izvorski 1997: 228, ex. 13)

(16b) can follow (16a) felicitously, indicating that the speaker is certain that John drank all the wine but does not have any evidence (although the speaker still needs some reason to believe 17a, as shown by the infelicity of 16c). Unlike (16), however, (17b) cannot be uttered felicitously after uttering (17a). According to Izvorski, this is due to the semantics of PE. Since PE in (17a) introduces the implication that the speaker has indirect evidence, the speaker cannot then say that he/she has none. ! 17

Based on these facts, Izvorski (1997) proposes that the semantics of the evidential marker EV with

the prejacent proposition p is as in (18).

(18)

The Interpretation of PE(p): a. Assertion: !p in view of the speaker’s knowledge state b. Presupposition: Speaker has indirect evidence for p (Izvorski 1997: 226, ex. 8)

In terms of Kratzer’s (1977, 1991) semantics of modals, (18) can be rephrased as follows: Just like English must, PE universally quantifies over possible worlds in the modal base: PE(p) is true if and only if in every epistemic alternative in the speaker’s state of knowledge, the prejacent p is true.17 However, unlike epistemic modals, the PE triggers the presupposition that the speaker has indirect evidence for p. This presupposition restricts the speaker’s state of knowledge, as PE introduces the felicity condition that the speaker’s state of knowledge should only contain the possible worlds where the speaker’s indirect evidence for p holds. In Chapters 3 and 4, I will discuss the issue of whether Korean evidentials, the direct -teand the reportative -ta-, behave in parallel with Izvorski’s (1997) PEs, exhibiting the characteristics of epistemic modals and introducing evidential presuppositions.

1.2.2. Illocutionary meaning: Faller (2002)

The second camp treats narrow evidentiality as being encoded in illocutionary operators and, thus, changing the illocutionary force of a sentence. In this proposal, evidentials do not !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 17

For convenience, I simplify the semantics of epistemic modals proposed in Izvorski (1997), which adopts a Kratzerian modal semantics.

! 18

contribute to the truth-conditions of a sentence (unlike epistemic modals), and the implications they introduce are different from presuppositions (e.g., Faller 2002).18 In this thesis, I review the most representative proposal, namely, Faller (2002). While analyzing evidential markers in Cuzco Quechua, Faller (2002) provides diagnostics to show that Cuzco Quechua evidential markers behave differently from epistemic modals, such as PEs. One diagnostics is whether an evidential marker can felicitously be used when the truth or falsity of the prejacent is already known. As the following English examples show, epistemic modals cannot felicitously be used when the speaker already knows the truth or falsity of the prejacent.

(19)

#It may/must be raining, but it is not (raining). (Faller 2002: 193, ex. 156)

(20)

#It may/must be raining, and it is raining.

However, Cuzco Quechua evidentials can be used even when the speaker already knows the truth or falsity of the prejacent, as in (21), where the reportative evidential -si- can be used felicitously even though the speaker knows that the prejacent (“they left me money”) is false.

(21)

Pay-kuna-s

ñoqa-man-qa

qulqi-ta

saqiy-wa-n,

mana-má

(s)he-PL-si

I-ILLA-TOP

money-ACC

leave-1O-3

not-SURP

ni

un

sol-ta

saqi-sha-wa-n-chu

not

one

Sol-ACC

leave-prog-1O-3-NEG

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 18

Matthewson et al. (2007) classifies Davis et al. (2007) as belonging to this camp, but strictly speaking, Davis et al. (2007) do not make any commitment to the semantic nature of the narrow evidentiality. They only discuss the pragmatic role of the sentence with evidential markers, no matter what the nature of the evidential implication might be.

! 19

‘They left me money, but they didn’t leave me one sol.’ Evidential implication: It is said / They said that they left me money. (Faller 2002: 193, ex. 155)

(21) shows that Cuzco Quechua evidential -si- behaves differently from epistemic modals and, therefore, is not an epistemic modal.19 To analyze evidentials in Cuzco Quechua, Faller (2002) claims that the semantics of evidential markers in Cuzco Quechua should be reduced to that of illocutionary operators, which affect the illocutionary force F (Searle 1969, Searle and Vanderveken 1985). An illocutionary force F, according to Faller (2002) and Searle and Vanderveken (1985), consists of various factors, including the illocutionary points and sincerity conditions.20 Illocutionary points indicate “how the propositional content of the illocutionary act relates to the world” (Faller 2002: 15). These are six illocutionary points: assertive, commissive, directive, declarative, expressive, and presentative. Sincerity conditions are “propositional attitudes of the form m(P), where m is a psychological mode such as, for example, desire, regret, or hope […] A performance of an illocutionary act is sincere when the speaker has the mental state that he expresses in the performance of that act, and it is insincere otherwise” (Vanderveken 1990: 117; quoted in Faller 2002: 17, emphasis from the original author). Faller (2002) claims that Cuzco Quechua evidentials are illocutionary operators rather than epistemic modals and formalizes them by adopting Searle and Vanderveken’s (1995) illocutionary logic and the definitions of the illocutionary force illustrated above. For example, !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 19

This example shows only that the operator being tested is not an epistemic modal, but not that it is an illocutionary operator. However, most of Faller’s (2002) tests focus on distinguishing propositional meaning from non-propositional meaning, and she seems not to consider other possibilities except epistemic modals introducing presuppositions and illocutionary operators. I do not pursue this issue further, but follow her reasoning for the moment. 20 There are other factors in the illocutionary force F, including the degree of strength and the mode of achievement, but I only indicate what is required for understanding the discussion below. For more detailed discussion, see Searle and Vanderveken (1985) and Faller (2002: Section 1.2.2).

! 20

Faller represents the lexical contribution of the Cuzco Quechua reportative evidential -si as in (22):

(22)

-si:

ASSERT SINC

(p)

->

= {Bel(s,p)}

PRESENT SINC =

(p)

{!s2[Assert(s2, p) " s2 #{h,s}]} (Faller 2002: 200, ex. 167)

According to (22), -si applies to the illocutionary point ASSERTION and changes it to PRESENTATION,

by which she means that the speaker does not make any commitment to the truth

or falsity of the prejacent p, but simply presents someone else’s assertion, like a quotation. Accordingly, the sincerity condition should also be changed. In assertions, the speaker should believe the truth of what is asserted, but in presentations, there should be another speaker who asserted the prejacent p who is neither the current speaker nor the current addressee. Similarly, the evidential -mi (indicating that the speaker’s assertion is based on the most reliable evidence available to the speaker) is illustrated in (23) and can be formalized as in (24):

(23)

Para-sha-n-mi. rain-Prog-3-mi p=’It is raining.’ Evidential implication: speaker sees that p21

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 21

In (24) the implication is that the speaker sees that it is raining and, therefore, following previous approaches to Cuzco Quechua, one might conclude that -mi is a direct perception evidential, but Faller (2002) claims that it indicates that the assertion is based on the speaker’s most reliable evidence, no matter what the type might be.

! 21

(24)

-mi:

ASSERT SINC

(p)

->

= {Bel(s,p)}

ASSERT (p) SINC =

{Bel(s,p), Bpg (s, Bel(s,p))} (Faller 2002: 167, ex. 130)

According to (25), -mi only changes the sincerity condition of the utterance, without changing the illocutionary force of the prejacent (unlike -si). Thus, it indicates that the assertion is based on the speaker’s best information (which may be direct perception, report, or inference). The Cuzco Quechua inferential evidential chá behaves differently from other evidentials, in the sense that it is an epistemic modal. An example of chá is given in (25):

(25)

Pilar-qa

yachay

wasi-pi-chá

ka-sha-n.

Pilar-TOP

know

house-LOC-chá be-PROG-3

‘Pilar must/may be at school.’ ev = speaker conjectures that Pilar must/may be at school

(elicited) (Faller 2002: 21, ex. 12a)

According to Faller, chá is an inferential evidential that introduces the implication that the speaker’s assertion is based on reasoning. Furthermore, since chá is an epistemic modal, it indicates that the prejacent is only a possibility and not an absolute fact (that is, by chá the speaker may believe that a prejacent is highly possible or may be less certain but does not believe that it is certain). For example, a sentence with chá cannot be followed by its negation, as in (26), just like the English epistemic modals in (27).

! 22

(26)

#Llave-qa key-TOP

muchila-y-pi-chá

kasha-n,

ichaqa

mana-n

backpack-1-LOC-chá

be-PROG-3,

but

not-mi

aqhay-pi-chu there-LOC-NEG ‘The keys may be / are possibly / probably in my backpack, but they are not there.’ (Faller 2002: 178, ex. 138)

(27)

#It is possible that God is alive, but he is dead. (Faller 2002: 178, ex. 139)

This claim is further supported by the fact that sentences with chá make weaker assertions than sentences without chá, as do epistemic modals. Consider (28):

(28) a. Para-sha-n-chá rain-PROG-3-chá ‘It is raining’ EV

= the speaker conjectures that it is raining

b. Para-sha-n rain-PROG-3 ‘It is raining’ EV

= the speaker sees that it is raining (Faller 2002: 178, ex. 140)22

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 22

Without any evidential marker, a sentence in Cuzco Quechua is interpreted as having the evidential implication that the speaker has direct (perceptive) evidence (Faller 2002: Ch. 1).

! 23

According to Faller, in (28b), the speaker is asserting that it is raining, whereas in (28a) the speaker only indicates that there is a possibility that it is raining. In this respect, as well, chá parallels epistemic modals. To capture epistemic modality and its evidential implications and to preserve the parallelism between chá and other two evidentials (which Faller analyzes as illocutionary operators), Faller (2002) proposes the following semantics for chá:

(29)

-chá:

ASSERT SINC

(p)

= {Bel(s,p)}

->

ASSERT ($p) SINC =

{Bel(s, $p), Rea (s, Bel(s, $p))} (Faller 2002: 167, ex. 130)

(29) says that chá affects both the assertion and the sincerity condition. In terms of assertion, by using chá, the speaker indicates that the prejacent is a possibility (and not a fact). In terms of sincerity conditions, the speaker indicates that the belief that the prejacent is possible is based on reasoning.23 Before moving on to the next section, note that most of all, the semantic content of these operators and the level where they are sensitive is completely unclear. This becomes an issue when we consider evidentiality in questions. As we will see later on, to account for evidentials in questions, Faller (2002) should allow two or more illocutionary operators to apply to a proposition at the same time. Specifically, Faller assumes that to ask a question is to request the addressee’s assertion, so she posits another illocutionary operator, REQUEST. Thus, evidentials in questions could be analyzed in terms of the ‘staking’ of the illocutionary operators EVI(DENTIAL) !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 23

Note that Faller still treats the epistemic modality as truth-conditional, in the sense that it directly affects the assertion; only the evidential implication is illocutionary. Thus, the question becomes whether there exists any evidential system in a natural language where an evidential marker is an epistemic modal and both the epistemic modality and the evidential implication are illocutionary (or at least non-truth-conditional). At the moment, I do not have an answer to this question.

! 24

and REQUEST. However, what would be the definition of REQUEST? And if two or more illocutionary operators were applied to a proposition at the same time, how would the illocutionary force of the sentence be computed? Again, I do not find any clear answer to this question in Faller (2002). I return to Faller’s (2002) proposal in Chapter 4 and consider in more detail whether it can be applied to Korean evidentials and the questions raised in this thesis.

1.2.3. Evidentials and tense morphemes

Besides epistemic modals, it is often observed that evidentials appear in the tense systems of several languages. For example, Faller observes that a certain type of tense morpheme may introduce evidentiality, even though these morphemes are not evidential markers by themselves. The example comes from the Cuzco Quechua tense morphemes -rqa and -sqa.

(30) a. Para-sha-sqa. rain-PROG-NX.PST24 ‘It was raining.’ EV

= speaker was told/infers that it was raining

b. Para-sha-rqa-n rain-PROG-PST-3 ‘It was raining.’ implicated EV = speaker saw that it was raining. (Faller 2004: 46, ex. 1)

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 24

NX.PST

= non-experienced past.

! 25

In (30a), the speaker is asserting that it was raining before the time of utterance, based on what someone else said or something inferred to have resulted from rain. Similarly, in (30b) the speaker is asserting that it was raining before the time of utterance, based on visual evidence. From these interpretations, these two morphemes introduce implications similar to those of evidentials. However, Faller claims that these morphemes do not encode any evidential value in their lexical semantics and only ‘implicate’ evidential meanings. If they were evidentials, they should not co-occur with other evidential markers, but this prediction is not borne out, as shown in (31).

(31) a. Para-sha-sqa-n. rain-PROG-NX.PAS-BPG ‘It rained.’ EV

= the speaker was told/infers from results that it rained.

b. Para-sha-sqa-s rain-PROG-NX.PST-REP ‘It rained.’ EV

= the speaker was told that it rained. (Faller 2004: 56, ex. 13)

Thus, the best-possible-grounds evidential mi (which appears as n in 31a) and the reportative evidential si co-occur with sqa in (31a) and (31b), respectively. Faller (2004) proposes that these are deictic tense morphemes, whose meaning is defined in terms of two spatiotemporal trace functions: e-trace and P-trace. First, e-trace takes an eventuality and pairs each time interval t in the temporal trace of that eventuality ($(e)) with the

! 26

location l. The output of e-trace is a set of time-location pairs , the ‘run-time-space’ of the eventuality.

(32)

e-trace(e) = {|t%$(e) " AT(e,t,l)} (Where AT(e,t,l) is true iff the eventuality e takes place at location l at time t.) (Faller 2004: 70, ex. 36)

Second, p-trace takes a speaker sp and pairs each time interval t in the temporal trace of that speaker ($(sp)) with the location l. The output of P-trace is also a set of time-location pairs , the ‘run-time-space’ where the speaker is located.

(33)

P-trace(e) = {|t%$(sp) " PERCEIVE(sp,t,l)} (Where PERCEIVE(sp,t,l) is true iff the speaker sp perceives location l at time t.) (Faller 2004: 70, ex. 37)

Roughly speaking, e-trace indicates the time and place where an eventuality occurs, and P-trace indicates the time and place where a speaker perceives. Given this, the deictic tense marker -sqa, which introduces the indirect evidential implication, is defined as in (35).

(34)

Meaning of -sqa: -sqa: %tR.%P.%e. P(e) " tR [t%tR " ' e-trace(e) -> ' P-trace(sp)] (Faller 2004: 71, ex. 38)25

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 25

Faller (2004) later revises this lexical entry, but this preliminary version is sufficient for our purposes.

! 27

That is, -sqa takes some past reference time interval tR, a proposition P, and an eventuality e and asserts that there is some time-location pair in e-trace in the past that does not overlap with P-trace. In other words, there is some time-location pair at which the event described in the proposition occurred but which the speaker did not perceive (or, very roughly speaking, there is some ‘part’ of an eventuality, which the speaker did not perceive). This means that, when using -sqa, the speaker should ‘infer’ the entire eventuality based on what was perceived. Hence, the indirect evidentiality arises with -sqa. What, then, is the nature of the indirect evidential implication introduced by -sqa? Faller (2004) says that -sqa ‘implicates’ the evidential meaning but remains unclear whether this means that the evidential implication of -sqa is a conversational implicature, just a presupposition, or something else entirely. Faller argues, however, that unlike other evidentials in Cuzco Quechua, which she analyzes as illocutionary operators, deictic tense markers operate at the level of propositions. This suggests that the evidential implication of -sqa is some kind of implicature: sqa is actually a tense marker and not an evidential, but it introduces evidential implications either as conventional or conversational implicatures. Chung (2005, 2007) extends Faller’s (2004) analysis of Cuzco Quechua to Korean, her main concern being the non-final ending -te-. Several researchers (e.g., Sohn 1994, Cinque 1999) have analyzed -te- as a direct evidential, as in (36a), but if it is a direct evidential, indicating that the speaker has direct evidence about the prejacent, then it remains unclear why it apparently introduces the inferential evidential implication when used with the future modal/tense marker keyss-, as in (35b).

(35) a. Ecey yesterday

pi-ka

o-te-la.

rain-Nom

fall-te-Decl

! 28

‘Yesterday it rained.’ Implication: the speaker saw it rain yesterday. b. Nayil tomorrow

pi-ka

o-keyss-te-la.

rain-Nom

fall-will-te-Decl

‘Tomorrow it will rain.’ Implication: the speaker infers that it will rain tomorrow.

Adopting Faller’s (2004) P-trace, Chung (2007) defines -te- as in (36):

(36)

Spatial deictic past tense -te [[-te-]]c is only defined if c provides a unique salient spatiotemporal location Lc such that $(Lc) < tc " Lc % P-trace(sc). If defined, then [[-te-]]c = Lc. (where tc = the speech time, sc = the speaker of the context c) (Chung 2007: 204, ex. 26b)

According to (37), -te- is only defined if the time interval is located in the past and the spatiotemporal location is included in the speaker’s perceptual trace. Under this analysis, when -te- is used without any overt modal/tense marker (as in 35a), the direct evidential implication is introduced as a presupposition - a definedness condition in the lexical entry in (36). The next question, then, is why we apparently have the inferential evidential implication in (35b). Chung’s (2005, 2007) answer is that it comes not from -te-, but from -keyss-, which, in Chung’s (2005, 2007) analysis, is ambiguous between an inferential evidential marker

! 29

and a modal/tense marker.26 I return to Chung’s (2005, 2007) analysis in a later chapter, but one problem with it is that we need to assume the unwelcome ambiguity of the modal marker -keyss-, such that it is an indirect evidential when the deictic tense marker -te- is present and a mood/modal marker when -te- is absent. If -te- marks deictic tense, why are the evidential meanings that are correlated with the presence or absence of -te- in complementary distribution? In any case, it seems that in this kind of analysis, the markers introducing evidential implications are not true evidentials. This parallels quotative constructions, which introduce implications similar to quotative / reportative evidential implications, but are not themselves evidentials. In other words, they express evidentiality in the broad sense, but not in the narrow sense. Therefore, I do not consider this third alternative in the following discussion, focusing instead on the first two.

1.2.4. Interim summary

In this subsection, I reviewed previous proposals on the linguistic category of evidentials, which have treated them as epistemic modals, as illocutionary operators, and as tense morphemes. When evidentials are epistemic modals, they introduce the evidentiality as presuppositions, whereas when they are illocutionary operators, they change the illocutionary force of the sentence hosting them. In some languages, they might be realized as tense markers, but I argue that the evidentiality expressed by specific tense morphemes is of the broad, rather than the narrow, type. In the rest of this dissertation, I argue, as mentioned before, that Korean evidential markers, direct -te- and reportative -ta-, are presupposition triggers, like PEs in Izvorski (1997) and St’át’imcets evidentials in Matthewson et al. (2007): they are partial identity functions, !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 26

I will return to Chung’s (2005, 2007) analysis in a later chapter and argue that contrary to what she claims, -te- can be analyzed as a direct evidential without assuming the lexical ambiguity of -keyss-.

! 30

introducing the evidential presuppositions.27 However, I will also claim that, unlike PEs or St’át’imcets evidentials, Korean evidentials do not contain any epistemic modal components (or, at least, there is no evidence enough to prove that there is any epistemic modal component in the lexical entry for Korean evidenials). Before moving on to the next chapter, it is essential to illustrate the similarities and differences between evidentials and indexicals. In the following subsections, I briefly review the semantics of indexicals and interrogatives. I postpone detailed introduction of the formal analysis of indexicals and questions until Chapter 4, where I introduce the issue this thesis deals with in more detail and make my own proposal.

1.3. Semantics of indexicals

As mentioned before, one characteristic aspect of semantic implications introduced by evidentials is that they vary depending on who the speaker is. In other words, they exhibit speaker-dependency (or speaker-anchoring). For instance, consider the following example of the Tibetan evidential (Garrett 2001: 228, ex. 5):

(37)

bkra.shis

za.khang-la

phyin-song

Tashi

restaurant-Loc

go-Dir.Past

‘Tashi went to the restaurant.’ Implication: the speaker has direct evidence that Tashi went to the restaurant.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 27

There are actually slight differences between -te- and -ta-, in that the former also seems to encode past tense meaning. I will return to this issue in Chapter 2 and 4.

! 31

If Mary utters (37), the implication introduced by song is that Mary has direct evidence that Tashi went to the restaurant. However, if Bill utters (37), then the implication changes accordingly—it is Bill who has direct evidence that Tashi went to the restaurant. In this sense, the implication of song is speaker-dependent; it changes depending upon the utterer.28 In fact, evidential markers are generally interpreted with respect to the speaker, regardless of the specific evidential system or type of evidentiality. For example, the implication introduced by the Japanese inferential evidential marker soo (e.g., McCready and Ogata 2007) also varies according to who the speaker is, such that if John utters (38), the implication is that John’s guess/inference is that it will rain, whereas if Mary utters (38), the implication is that Mary’s guess/inference is that it will rain.

(38)

gogo-kara

ame-ga

furi-soo-da.

Afternoon-from

rain-Nom

come-soo-Cop.Decl

‘It will rain this afternoon.’ Implication: Based on the speaker’s guess/inference, it will rain. (Emi Mukai, p.c.)

Compare this kind of implication with the presupposition introduced by the English focus particle also.

(39)

[John]f also went to the restaurant.29 Presupposition: there is a person different from John who went to the restaurant.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 28

Throughout this thesis, I use the terms speaker-dependent and anchored to the speaker to refer to this characteristic of evidential implications. 29 By [(]f I mean that ( associates with an intonational focus.

! 32

Very roughly speaking, the focus particle also introduces the presupposition that the there is another contextually salient alternative different from the focused element and that the proposition where the focused element is replaced by this salient alternative is also true.30 Unlike the implication introduced by evidentials, this presupposition is speaker-independent. Thus, regardless of who utters (39), the presupposition it carries remains the same. The speaker-dependency of the evidential implication is reminiscent of the interpretation of indexicals, which are also speaker-dependent. Consider the following example:

(40)

I am rich.

The truth-condition of this sentence varies depending on the speaker. If John utters (40), then it is true if and only if John is rich; if Mary utters (40), then it is true if and only if Mary is rich, and so forth. This type of speaker-dependency is due to the interpretation of the first-person pronoun I, which always denotes the utterer. Several expressions in natural language exhibit this type of speaker-dependency. For example, you always denotes the addressee; here denotes the place where the speaker utters it; now denotes the time when the speaker utters it, etc. Kaplan (1989) calls such expressions indexicals (Kaplan 1989: 490).31 They are assigned their denotations directly from speaker’s context, without any intermediate level of meaning such as Fregean sense or intension. In addition to pronouns and certain adverbials, several expressions in natural language also exhibit indexicality, including evidentials. Some researchers have observed that expressives (damn, jerk, etc.; see, e.g., Potts 2005, 2007) and utterance-modifying adverbs (honestly, frankly, etc.; see, e.g., Potts 2005, Ernst 2009) are also speaker-dependent. Some examples appear in (42). !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 30 31

For a detailed introduction to the semantics of focus, see Rooth (1985, 1996). Kaplan attributes this term to Charles Sanders Peirce but seems unsure in doing so.

! 33

(41) a. John is planning to quit that damn job. Implication: the speaker believes that John’s job is pretty bad. b. Honestly, John does not know that. Implication: the speaker utters honestly (without any deception) c. Purportedly, John quit the job. Implication: the speaker heard from someone else that John quit the job.

There are exceptions, of course. Speaker-dependency may vary or even not apply in different languages. In English, for example, the denotations of pronouns are always speakerdependent, as mentioned above, but Schlenker (2003) observes that in languages such as Amharic, the denotation of pronouns such as I may not be speaker-dependent under the scope of an attitude operator, as shown in (42).

(42)

Situation to be reported: John says: ‘I am a hero’. a. Amharic (lit.): Johni says that Ii am a hero. b. English:

Johni says that hei is a hero /*Johni says that Ii am a hero. (Schlenker 2003: 31, ex. 3)

Anand and Nevins (2004) also report similar cases in Zazaki (an Indo-Iranian language spoken in eastern Turkey) and Slave (an Athabaskan language spoken in northwestern Canada), where indexicals are shifted under verbs of saying. Language is not the only parameter affecting the speaker-dependency of expressions such as indexicals and expressives. Conversational background may also change speaker-dependency. In the case of expressives, Amaral et al. (2007) and Harris and Potts (2010) propose that they ! 34

need not always be speaker-dependent, as in (43), where the expressive friggin may not be anchored to the speaker.

(43)

Given a context where we know that Bob loves to do yard work and that his son Monty hates to, Bob could say the following (perhaps in response to a partner’s suggestion that Monty be asked to mow the lawn while Bob is away on business): Well, in fact Monty said to me this very morning that he hates to mow the friggin lawn. (Harris and Potts 2010: 6)

These exceptions are very limited, and are based on the specific linguistic and/or extra-linguistic context. For example, one might think that the exceptional shift illustrated above is due to the specific predicate (that is, verbs of saying) or some special pragmatic context. A full investigation of the conditions affecting speaker-dependency is, of course, beyond the scope of this thesis, so I leave these questions open. However, there is one exceptional case that occurs more regularly and shows the distinction between evidentials and other indexicals more clearly. That is, cases where evidentials are used in questions. Consider the examples in (44).

(44) a. John-i John-Nom

na-lul

po-te-la.

I-Acc

see-te-Decl

‘John saw me’ Implication: the speaker saw that John saw the speaker b. John-i John-Nom

na-lul

po-te-nya?

I-Acc

see-te-Q

! 35

‘Did John see me?’ Implication: the addressee is expected to know whether John saw the speaker or not

Both in declaratives, such as (44a), and in questions, such as (44b), the first person pronoun na ‘I’ refers to the speaker. The evidential implication, however, behaves differently. In declaratives, it is anchored to the speaker, but in questions, it is anchored to the addressee, as shown in (44a) and (44b), respectively. Why do evidentials and other indexicals behave differently? Furthermore, unlike the cases in (42) and (43), in questions, evidentials shift from the speaker’s perspective to the addressee’s. How can we account for this obligatory shift? One way to address this problem is to distinguish the notion of asserter from the notion of utterer. When a speaker utters a declarative sentence, he/she also asserts a proposition denoted by that declarative sentence. However, when a speaker utters a question, he/she does not assert any proposition, but only the addressee asserts a proposition as an answer to the questioner. We also saw that, the evidential is anchored to the utterer in declaratives and to the addressee in questions, whereas other indexicals are anchored to the utterer both in declaratives and in questions. Therefore, the difference between evidentials and other indexicals can be formulated as follows: evidentials are always anchored to the asserter, whereas other indexicals are always anchored to the utterer.32 Then our next question is how to capture the asserter-dependency of evidentials, which is different from the utterer-dependency of other indexicals. In Chapter 4, I make my own proposal, which accounts for this phenomenon without major modifications to previous theories of evidentials, interrogatives, or indexicals, and provides further insights into the semantics of indexicality and evidentiality more broadly.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 32

Thanks to Stephen Finlay for pointing out this difference.

! 36

1.4. Conclusion and outline of the dissertation

In this chapter, I first reviewed previous approaches to the semantics of evidentiality. We saw that the implications introduced by evidentials are analyzed as either presuppositions or illocutionary acts and, accordingly, that evidentials are treated as either epistemic modals or illocutionary operators. In some languages, tense operators introduce implications similar to evidential implications, but I argued that this does not mean that tense operators should be analyzed in terms of evidentiality. I also outlined the standard theory of indexicals and interrogatives. I adopted Kaplan’s (1989) theory of indexicals, assuming they receive their denotations directly from the speaker’s context. I also showed that there are several other expressions which are speaker-dependent, such as expressives, utterance-modifying adverbs, and evidentials, but that their speaker-dependency is not absolute and may vary depending on language, context, and construction. In the following chapters, I investigate evidentiality in Korean and its interactions with interrogatives. To do so, it is important to establish the evidential system of Korean, which I do in the next chapter. In Chapter 2, I address the question of whether there is evidentiality in Korean and, if there is, what type is grammatically realized. I argue, based on morphological, syntactic, and semantic considerations, that there are two types of evidential morphemes in Korean: the direct evidential -te- and the reportative evidential -ta-. Many scholars agree that -te- is a direct evidential morpheme or at least is responsible for introducing a kind of evidentiality when combined with other operators, such as tense morphemes or aspect morphemes. What remains controversial is whether there is a reportative evidential morpheme in Korean. I focus primarily on the question regarding the Korean reportative evidential and argue that there is a reportative evidential morpheme in Korean. Then, in Chapter 3, I discuss the semantics of evidentials in ! 37

Korean, showing that the semantic implication introduced by Korean evidential morphemes is a presupposition, but evidence presented in the previous literature to support the claim that Korean evidentials are epistemic modals is not much compelling. Chapter 4 presents the main proposal of this thesis. There, I illustrate the puzzle I call the evidential perspective shift and show that previous approaches to similar phenomena cannot be extended to Korean. I also argue that this puzzle should be addressed in semantic terms and that a purely pragmatic approach does not work for the Korean facts. Finally, I propose that this question can best be answered by assuming that Korean evidentials are functions from propositions to characters. Chapter 5 concludes the thesis and summarizes the remaining problems, including a discussion of other questions raised throughout the thesis and some speculations on possible answers.

1.5. Further discussion 1: interrogatives

Since I focus on the interaction between evidentials and interrogatives, it is worth reviewing the semantics of interrogatives that I adopt in this thesis. There are two major approaches to the semantics of interrogatives. Hamblin (1973) and Karttunen (1977) analyze questions as sets of propositions representing possible instantial answers, whereas Groenendijk and Stokhof (1984) analyze them as partitions over possible worlds.33 The two analyses make different empirical predictions, especially concerning embedded questions. In this thesis, I adopt the former and assume that questions are sets of answers because this greatly simplifies the

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 33

For a detailed introduction to the semantics of questions, see, e.g., Ginzburg (1996), Guerzoni (2003: Ch. 1), and Higginbotham (1996).

! 38

computation and because, as we will see later, Korean evidentials are not generally embedded.34 For example, the question in (45a) has the extension in (45b), when the domain of individuals includes John, Mary, and Bill. A speaker asks the question in (45a) as a set of possible instantial answers, and an answerer is expected to select one or more of the answers that the speaker believes are true.

(45) a. Who left? b. {%w. John left in w, %w. Mary left in w, %w. Bill left in w}

Both Hamblin’s (1973) and Karttunen’s (1977) approaches analyze questions as sets of answers, but they differ in two respects. First, Hamblin (1973) assumes that a question is the set of all (possible) answers, whereas Karttunen (1977) assumes that it is the set of all true answers. Karttunen (1977) notes that is unclear how these two assumptions differ empirically but predicts that there may be differences when questions are embedded. However, since we focus on evidentials, which generally cannot be embedded, I ignore this difference for the moment. The second difference concerns how the extension of a question is derived. In Hamblin (1973), it is derived via a lexical entry for wh-words and a set-tolerant functional application rule. Specifically, he assumes that wh-words are sets of individual concepts in the domain. For example, the lexical entry for who is (46) (where there are three people in the domain, as assumed before).

(46)

[[who]] = { %w. John, %w. Bill, %w. Mary}

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 34

Later, I will provide a least tentative explanation for why evidentials generally cannot be embedded.

! 39

From this lexical entry, the extension of the question in (45a) is derived directly from the following LF:

Figure 1.2. LF of (45a) Who left? in Hamblin (1973) qp who left Like Hamblin (1973), Karttunen (1977) treats wh-words as regular quantifiers, but unlike Hamblin (1973), instead of deriving the set of answers directly from the lexical entries of whwords and set-tolerant functional application rules, he assumes a covert question morpheme, which I refer to as ?, which allows the shift from a proposition to a set thereof. In other words, it takes as an argument a proposition and returns a singleton whose only element is the original proposition (therefore, it is of type >).

(47)

[[?]] = %pst.{p}

To derive a denotation of a question from quantifiers (that is wh-words) and Qmorpheme, Karttunen (1977) further assumes wh-quantification rule. However, to avoid assuming any additional rule, we may revise the lexical entry for wh-words. For convenience, if we assume that who is the same as which person, then which should have the denotation in (48), and which person (=who) should have the denotation in (49).

(48)

[[which]] = %fet. %P>. %pst. !x [f(x)=1 " P(x)(p)=1]

(49)

[[which person]] = %P>. %pst. !x [person(x)=1 " P(x)(p)=1]

! 40

Since wh-words in this analysis are quantifiers over sets of propositions, they should undergo movement like other quantifiers and leave traces of type e. Therefore, in Karttunen (1977), the LF of (46a) is as in Figure 1.3.

Figure 1.3. LF of (45a) Who left? in Karttunen (1977) wo 3 3 which person1 1 3 ? 3 t1 left In most cases, as I mentioned before, it is difficult to find any differences in the empirical predictions of Hamblin’s (1973) and Karttunen’s (1977) semantics of questions, except those concerning embedded questions and multiple wh-questions. However, in a later chapter we will see that, these two variations also differ in their accounts of the interaction between evidentials and interrogatives. A detailed formal introduction is provided in Chapter 4.

1.6. Further discussion 2: an overview of evidential scales

Evidentiality is usually said to encode two different linguistic notions: the type of evidence and the degree of certainty of the speaker. Even though these two notions are distinguished logically, it is claimed that they are closely related and intertwined and, therefore, that we need to consider both dimensions in order to identify evidential markers in a given language or to classify different languages based on their evidential systems. However, I only provide a brief overview of the problem of evidential scales in this section, because i) the evidential scale is not particularly relevant to the main issues dealt with in this thesis, ii) degree of certainty as connected to type of evidence is a more complicated issue than type of evidence by ! 41

itself, and iii) as we will see later, it is not clear whether the problem of the evidential scale is linguistically significant or only a matter of world knowledge and not directly related to the nature of language. To discuss the scale of evidentiality, let us begin by reviewing Willett’s (1998) proposal about the types of evidence available in narrow evidentiality.

Figure 1.4. Types of information sources in Willett (1988) (repeated from above) Types of Evidence q p Direct | Attested e|i visual auditory sensory

Indirect q p Inferred Reportative ei e | i result-based reasoning-based second-hand third-hand folklore (from Lee J. 2010)

Willett’s (1988) taxonomy in Figure 1.4. makes several predictions about the types of evidence that can be grammaticalized in natural languages, among which I would like to mention two as concrete examples. First, the evidential system in a language is determined by how that language divides the hierarchy. Thus, one language may only distinguish direct (firsthand) evidentiality from indirect (non-firsthand) evidentiality, while another language may distinguish direct, inferred, and reportative evidentiality. Second, no language can grammaticalize evidentialities that belong to different ‘mother nodes.’ Thus, no natural language can have an evidential marker that indicates that the speaker has either visual or second-hand evidence, since visual evidence is a type of direct evidentiality, whereas second-hand evidence is a type of reportative evidentiality. Willett’s (1986) taxonomy also makes predictions with respect to degree of certainty. The farther left on the scale in Figure 1.4, the more certain the speaker is about the truth of the ! 42

assertion. Therefore, in Willett’s hierarchy, we may say that direct evidentials express the highest degree of speaker certainty concerning the prejacent, that inferential evidentials express a lower degree, and that reportative evidentials express the lowest degree of certainty. Other researchers have proposed their own cross-linguistic evidential categories. de Haan (1998), for instance, proposes the following linear hierarchy:

(50)

VISUAL > AUDITORY > OTHER SENSORY > INFERENCE > SECONDHAND > THIRDHAND > ASSUMED

(de Haan 1998; quoted in Faller 2002: 65)

Although de Haan (1998) does not present a universal hierarchy (Faller 2002: 48), this hierarchy is similar to Willett’s. One should note, however, that the category ASSUMED is located below the category THIRDHAND. It is unclear what de Haan (1998) meant by ASSUMED, but in general, like Willett (1998), de Haan (1998) assumes that direct perceptive evidentials express more certainty than inferential evidentials and that reportative evidentials express the least certainty of all. Based on these observations, one might conclude that there are three major types of evidentiality in natural languages: direct perception, inference, and reportative. This is a bit hasty, however, as there are evidentials in some language that do not appear in the proposed inventory of evidentials. For example, the evidential system of Kashaya (a Pomoan language) makes a fiveway distinction—performative (based on the speaker’s behavior and/or performance), factual/visual, auditory, inferential, and quotative—and this obviously does not fit into any of the hierarchies or classifications proposed above. Cuzco Quechua -mi, a so-called direct perception evidential marker, can be better formulated as best possible evidence (Faller 2002), indicating that the speaker is asserting based on the best available evidence, regardless of what type of ! 43

evidence it might be. Even without discussing the empirical data that contradict the proposed evidential hierarchies, one can easily imagine cases where the abovementioned hierarchies make incorrect empirical predictions. In typical cases, evidence based on direct perception is the strongest. However, suppose that one makes an assertion based on a mathematical proof. That kind of evidence comes from inference, but since mathematical truths are usually regarded as transcendent and absolute one might regard that kind of evidence as stronger than evidence from direct perception. In some cases, even the reportative evidential may be regarded as indicating stronger evidence than direct perception. Suppose, for example, that the secondhand evidence comes from a reliable source, such as an encyclopedia. In this case, direct perceptive evidence can be less reliable than reportative/secondhand evidence. Such examples are the main reason Faller (2002) proposes that the cross-linguistic evidential hierarchy is ‘non-linear.’ Specifically she claims that it is impossible to order every type of evidential and, therefore, that we need to assume that there are two evidential ‘clines’ which hold cross-linguistically: the ‘personal evidence’ cline and the ‘mediated evidence’ cline.

(51) a. The Personal Evidence Cline PERFORMATIVE > VISUAL > AUDITORY > OTHER SENSORY > INFERENCE FROM RESULTS > REASONING > ASSUMPTION

b. The Mediated Evidence Cline DIRECT > SECONDHAND > THIRDHAND > HEARSAY / FOLKLORE

(Faller 2002: 70)

Faller (2002) does not provide concrete examples of how these two hierarchies interact to determine the evidential system of a language. Furthermore, in some languages, a single evidential can express different degrees of certainty, and it is unclear how this kind of variability ! 44

can be accounted for in approaches that assume a hierarchical structure of evidential markers, such as de Haan’s (1998). For example, Matthewson et al. (2007) and Rullmann et al. (2008) point out that evidentials in St’át’imcets express different degrees of speaker’s certainty depending on the context. Consider the following two examples from Matthewson et al. (2007: 236, ex. 78):

(52)

Context: The speaker is talking about when, as a child, she used to play in the evenings with her friends. a. na DET

s-pála7-s-a,

wá7-lhkalh

k’a

wenácw-ts-am’

NOM-one-3POSS-EXIS

IMPF-1PL.SUBJ

INFER

true-mouth-MID

láti7

nu

wa7

wá7,

DEIC

DET

IMPF

be

“One time, we must have been loud.” b. ni. . .lh

s-zaw’t-min-tumúlh-as

k’a

FOC

NOM-fed.up-RED-1PL.OBJ-3ERG INFER

s-Ernest

Jacob

NOM-Ernest

Jacob

“And Ernest Jacob, who was living there, got fed up with us.”

According to Matthewson et al. (2007), the inferential evidential k’a in (52a) expresses that the speaker is certain that she and her friends were very loud. Then, in (52b), when the speaker asserts that the person living there, Ernest Jacob, was angry at her and her friends based on her inference, k’a expresses the speaker’s strong certainty that Ernest Jacob was angry. However, consider the following example from Matthewson et al. (2007: 237, ex. 80):

(53)

cw7aoz

kw-en-wá

stexw lexláx-s

lh-as

NEG

DET-1SG.POSS-IMPF

very

HYP-3CONJ

remember-CAUS

! 45

kás-tum’

i

what-1PL.ERG DET.PL

sk’wílh-a

ts’úqwaz’

leftover-EXIS

fish

“I don’t remember what we did with the leftover fish.” wa7

k’a

qelh-n-ás

nilh

kw

s-ts’áqw-an’-em

IMPF

INFER

put.away-DIR-3ERG

FOC

DET

NOM-eat-DIR-1PL.ERG

lh-kalál-as hyp-soon-3conj “Maybe she put it away and we ate it later.”

Since the speaker explicitly mentions that not remembering what was done with the leftover fish, the inferential evidential k’a in the next sentence cannot express strong certainty. In this case, it only indicates that it is possible that someone else put it away—the speaker is not sure about that. (52) and (53) show that the St’át’imcets inferential evidential k’a does not encode the degree of a speaker’s certainty in its lexical semantics. Matthewson et al. (2007) and Rullmann et al. (2008) also show that two other evidential markers in St’át’imcets, ku7 and an, exhibit the variation in the degree of certainty expressed. More cross-linguistic studies are required, and I do not have any conclusion at the moment. I doubt, however, that there is any evidential hierarchy in natural languages. It seems to me that there are several types of evidence that can be linguistically realized as evidentials, but the degree of certainty is not linguistically encoded in evidential markers. Summarizing the discussion in this last section of the introduction, I have tentatively concluded that the differences in degree of certainty between different evidentials are extralinguistic. There is some ‘tendency’ in the correlation between the type of evidence available to

! 46

the speaker and the speaker’s degree of certainty, but this ‘tendency’ seems to be easily overridden by pragmatic factors. Therefore I do not discuss this issue any more in this thesis.

! 47

Chapter 2 Evidentiality in Korean

In this chapter, I illustrate how evidentiality is realized as a morphologically separate and independent category in Korean. Specifically, I argue that there are at least two evidentials in Korean: reportative -ta- and direct -te-. It is commonly argued that the Korean ending -te- is an evidential indicating that the speaker has direct evidence (e.g., Sohn 1994, Song 1997, 1998, Lee J. 2010), but the existence of a reportative evidential in Korean is still debated, with some researchers recognizing the sentential ending -tanta as a whole as a reportative evidential ending (e.g., Kim N.-K. 1990, 2001, Chung K.-S. 2005), but others treating the same ending as a phonological contraction of the embedding predicate that includes the verb of quotation and, thus, carries similar implications to a sentence with a reportative evidential marker (e.g., Chae 2006). Therefore, in this chapter, I mainly focus on -ta-, arguing that this morpheme is a reportative evidential marker and that it exhibits many distributional and semantic differences from quotations. This chapter is organized as follows. In Section 1, I discuss reportative evidentiality in Korean. First, I overview several sentential endings that introduce reportative evidentiality and abstract the pre-final ending -ta- as a Korean reportative evidential. Then, I show that sentential endings containing -ta- are semantically as well as morphologically distinct from the phonological contraction of a quotative predicate. Finally, I outline two studies, Kim N.-K. (1990) and Chung J.-Y. (2009), that make proposals regarding Korean reportative evidentiality that are different from mine, and present several pieces of evidence against their claims.

! 48

Section 2 deals with direct evidentiality. There, I adopt previous analyses on Korean direct evidentiality, including Sohn (1994), Cinque (1999), Song (1997, 1998), Lee J. (2010), among others, claiming that in Korean the pre-final ending -te- indicates that the speaker has direct evidence. I also argue that -te- is not a deictic tense morpheme, as is suggested of Cuzco Quechua -sqa (Faller 2004), contrary to Chung K.-S. (2005, 2007). In Section 3, I briefly review Kwon’s (2010) proposal that Korean also has inferential evidential morphemes. Section 4 summarizes the chapter.

2.1. Reportative evidentiality in Korean

In Korean, a number of sentence-final endings alternate between forms without non-final -ta- and forms with -ta-. The difference between these two is that only the latter conveys a reportative evidential implication. Compare the examples in (1), where there is no non-final -ta-, with those in (2), where non-final -ta- is present.

(1) a. John-i John-Nom

Bill-ul

ttayli-ess-ta.

Bill-Acc

beat-Past-Decl

Bill-ul

ttayli-ess-supnita.

Bill-Acc

beat-Past-Decl

‘John beat Bill’ b. John-i John-Nom ‘John beat Bill’

! 49

c. John-i John-Nom

Bill-ul

ttayli-ess-e (=> ttayli-ess-tay)

Bill-Acc

beat-Past-Decl

Bill-ul

ttayli-ess-ta-nta. 35

Bill-Acc

beat-Past-ta-Decl

‘John beat Bill’

(2) a. John-i John-Nom ‘John beat Bill’

Implication: the speaker heard that John beat Bill from someone else. b. John-i John-Nom

Bill-ul

ttayli-ess-ta-pnita.

Bill-Acc

beat-Past-ta-Decl

‘John beat Bill’ Implication: the speaker heard that John beat Bill from someone else. c. John-i John-Nom

Bill-ul

ttayli-ess-ta-e

Bill-Acc

beat-Past-ta-Decl

‘John beat Bill’ Implication: the speaker heard that John beat Bill from someone else.

The sentential endings in (1) include the tense marker -ess- followed by declarative markers (which vary depending on the register and the social status of the speaker and addressee) and do not contain any other (overt, at least) morphemes. These sentences do not carry any kind of evidential implication, that is, we do not know what kind of evidence the assertions in (1) are !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 35

In (2a), there is another marker -n- between two -tas, which is commonly analyzed as a present tense morpheme, but in -tapnita and -tay we do not observe any present tense morpheme corresponding -n-. Some linguists, such as Gim (2000), assume that there is a covert present-tense morpheme corresponding to -n- in -nta, but for convenience, we treat the entire -nta as an allomorph of the declarative marker -ta in (1a). Similarly, some authors, including descriptive linguists in Korean, also argue that the -pnita in -tapnita should be analyzed further, but for simplicity, we treat it as single declarative ending. These assumptions do not affect my main claim that there is a reportative evidential ending in Korean.

! 50

based on.36 These endings contrast with the sentential endings in (2), where -ta- precedes the declarative markers. As indicated in the translation, when the declarative endings are preceded by -ta-, the sentence carries the implication that the speaker has reportative evidence about what he/she is asserting (or that the speaker heard from someone else about what he/she is asserting). We find a similar contrast in questions. In (3), where the interrogative ending is simple, there is no evidential implication, but in (4), where the interrogative ending is preceded by -ta-, the question carries the implication that the speaker expects the addressee to answer on the basis of reportative evidence.

(3)

Question endings without -taa. John-i John-Nom

Bill-ul

ttayli-ess-ni?

Bill-Acc

beat-Past-Q

‘Did John beat Bill?’ b. John-i John-Nom

Bill-ul

ttayli-ess-supnikka?

Bill-Acc

beat-Past-Q

‘Did John beat Bill?’ c. John-i John-Nom

Bill-ul

ttayli-ess-e? (=> ttayli-ess-tay)

Bill-Acc

beat-Past-Q

‘Did John beat Bill’

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 36

Actually, these sentences carry the implication that the speaker’s assertion is based on his/her most reliable evidence, regardless of the type, but this implication can be accounted for in terms of Grice’s (1975) cooperative principle. I do not discuss the evidential implication in sentences without overt evidential markers (but see Chapter 5 for relevant discussion).

! 51

(4)

Question endings with -taa. John-i John-Nom

Bill-ul

ttayli-ess-ta-ni?

Bill-Acc

beat-Past-ta-Q

‘Did John beat Bill?’ Implication: the speaker expects that the addressee heard from someone else whether John beat Bill. b. John-i John-Nom

Bill-ul

ttayli-ess-ta-pnikka?

Bill-Acc

beat-Past-ta-Q

‘Did John beat Bill?’ Implication: the speaker expects that the addressee heard from someone else whether John beat Bill. c. John-i John-Nom

Bill-ul

ttayli-ess-ta-e? (=> ttayli-ess-tay)

Bill-Acc

beat-Past-ta-Q

‘Did John beat Bill?’ Implication: the speaker expects that the addressee heard from someone else whether John beat Bill.

For brevity, hereafter I refer to sentential endings without -ta- as ‘simple endings’ and to the sentential endings with -ta- as ‘complex endings.’ Given this, the paradigm of Korean sentential endings shown in (1-4) is as summarized in (5).

! 52

(5)

Simple

Complex

Declarative

Question

Declarative

Question

Formal Style

-pnita

-pnikka

-ta-pnita

-ta-pnikka

Plain Style

-(n)ta

-ni

-ta-nta

-ta-ni

-e/a

-ta-e/a (=> tay) -ta-e/a (=> tay)37

Intimate Style -e/a

2.1.1. The traditional view

In analyzing complex endings, many scholars adopt a position that stems from historical considerations, treating complex endings as phonological contractions of embedding matrix verbs involving verbs of quotation whose complement is a proposition.

(6)

-tanta / -tapnita / -tay

=>

-tako malha-nta / -tako malha-pnita / -tako malha-e.

-ta:

Declarative ending

-ko:

Complementizer

-mal:

Verb ‘say’

-nta /-pnita / -e:

Declarative ending

Strictly speaking, these constructions are not ‘quotations’ (so-called direct quotations), but constructions in which the embedding predicate involves a verb of saying (so-called indirect quotations). However, for convenience, I do not strictly distinguish constructions containing verbs of saying from quotations, and unless indicated otherwise, I use the term ‘quotation’ only to refer to indirect quotations, because this is the only form of quotation most relevant to the topic of this chapter. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 37

In the ‘intimate style,’ the declarative ending and the question ending take the same form, differing only in intonation.

! 53

For brevity, hereafter I refer to the analysis of complex endings shown in (6) as the traditional analysis. Under the traditional analysis, complex endings are derived via the following stages, all of which surface in Korean. First, in quotative constructions, the embedding subject can be omitted whenever it is contextually retrievable or generic, as shown in (7).

(7) a. John-i John-Nom

Bill-ul

ttayli-ess-ta-ko

malha-nta.

Bill-Acc

beat-Past-ta-Comp

say-Decl

‘(Someone says/They say) John beat Bill’ b. John-i John-Nom

Bill-ul

ttayli-ess-ta-ko

malha-pnita.

Bill-Acc

beat-Past-ta-Comp

say-Decl

‘(Someone says/They say) John beat Bill’ Implication: the speaker heard that John beat Bill from someone else. c. John-i John-Nom

Bill-ul

ttayli-ess-tako

malha-e

Bill-Acc

beat-Past-ta-Comp

say-Decl

‘(Someone says/They say) John beat Bill’

Secondly, the embedding predicate malha- ‘say’ is replaced by the pro-verb -ha, which is commonly translated ‘do.’

(8) a. John-i John-Nom b. John-i John-Nom c. John-i John-Nom

Bill-ul

ttayli-ess-ta(-ko)

ha-nta.

Bill-Acc

beat-Past-ta-Comp

do-Decl

Bill-ul

ttayli-ess-ta(-ko)

ha-pnita.

Bill-Acc

beat-Past-ta-Comp

do-Decl

Bill-ul

ttayli-ess-ta(-ko)

ha-e

Bill-Acc

beat-Past-ta-Comp

do-Decl ! 54

Thirdly, in colloquial Korean, the complementizer ending -ko is easily omitted, leaving the consonant [h] intervocalic, where [h] typically contracts into -ay.38 The result is illustrated in (9).

(9) a. John-i John-Nom b. John-i John-Nom c. John-i John-Nom

Bill-ul

ttayli-ess-ta-ay-nta.

Bill-Acc

beat-Past-ta-V-Decl

Bill-ul

ttayli-ess-ta-ay-pnita.

Bill-Acc

beat-Past-ta-V-Decl

Bill-ul

ttayli-ess-ta-ay-e

Bill-Acc

beat-Past-ta-V-Decl

(ttayli-ess-tay-nta)

(ttayli-ess-tay-pnita)

(=> ttayli-ess-tay).

Finally, as pointed out by Chung J.-Y. (2009) and Sohn (1999), among others, the [h] in ha- is typically elided when intervocalic, yielding the realizations in (10), which are the same as (2).

(10) a. John-i John-Nom b. John-i John-Nom c. John-i John-Nom

Bill-ul

ttayli-ess-ta-nta.

Bill-Acc

beat-Past-ta-Decl

Bill-ul

ttayli-ess-ta-pnita.

Bill-Acc

beat-Past-ta-Decl

Bill-ul

ttayli-ess-ta-e

Bill-Acc

beat-Past-ta-Decl

(=> ttayli-ess-tay).

Under the traditional analysis, (9) and (10) are treated as free variants of the same phonological contractions of the quotations. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 38

This is due to the irregular inflection of ha-, which I do not discuss further. Additionally, to keep things simple, I gloss ay simply as V.

! 55

In the next subsection, I argue that the traditional analysis is correct about the examples in (9) but wrong about those in (10), which feature instances of reportative evidential constructions rather than phonological contractions of embedding predicates involving the verbs of quotation in (7) and (8). To do so, in the following subsections, I argue that the examples in (10) are reportative evidential constructions, whereas the examples in (9) are phonological contractions of the quotative constructions in (7) and (8). In the remainder of this section, I provide a stricter set of semantic and distributional diagnostics and argue that Korean has a reportative evidential marker that should be treated separately from its quotative constructions. First, I discuss the semantic diagnostics and then turn to the distributional diagnostics.39

2.1.2. Semantic diagnostics

In this subsection, I discuss the following two semantic diagnostics, showing that -tanta behaves semantically differently from phonological contractions of embedding predicates containing the verb of quotation:

- Questions and negation - First person effects

2.1.2.1. Question and negation

My first argument is that -tanta behaves differently from embedding verbs of saying in questions and under negation. Let us first consider the case of questions. When the quotative

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 39

For convenience, hereafter I focus primarily on -tanta.

! 56

predicate is used in a question, the question may ‘target’ the quotative embedding predicate but not the embedded proposition. Consider (11).

(11)

Bill-un

Tom-i

na-lul

po-ass-ta-ko

malha-ni?

Bill-Top

Tom-Nom

I-Acc

see-Past-Decl-Comp

say-Q

‘Does Bill say that Tom saw me?’

Here, the question concerns whether Bill says that Tom saw the speaker, and not whether Tom actually saw the speaker. When the subject of the embedded clause is the same as that of the matrix clause, the embedded subject may be omitted. An example is given in (12).

(12)

Tom-i

na-lul

po-ass-ta-ko

malha-ni?

Tom-Nom

I-Acc

see-Past-Decl-Comp

say-Q

‘Does Tom say that he saw me?’

The point is that the question is still targeting the embedding predicate; (12) is asking whether Tom says something, not whether Tom saw the speaker. (13), where the contracted verb -ay- appears in the question, has the same interpretation as (12).

(13)

Tom-i

na-lul

po-ass-ta-ay-ni?

Tom-Nom

I-Acc

see-Past-Decl-V-Q

‘Does Tom say that he saw me?’

! 57

Compare the examples in (11) through (13) with (14), where -tani, the interrogative counterpart of -tanta, is used in the same question.

(14)

Tom-i

na-lul

po-ass-ta-ni?

Tom-Nom

I-Acc

see-Past-Q

‘Did Tom see me?’ Implication: Tom has reportative evidence that Tom saw the speaker.

Here, the target of the question is the prejacent,40 corresponding to the embedded proposition in (12): the speaker is asking whether Tom saw the speaker, not whether someone else does or does not say that Tom saw the speaker. That is, the reportative evidential implication in -tanta is not (and actually cannot be) the part of hat the question is intended to inquire about. The interpretation of -ta- in questions is also different from the quotation of a question. This can be further shown by the fact that, in order to embed a question, we need to adopt the phonological contraction of the pro-verb corresponding to the verb of the question. Compare (14) with the phonological contraction of the quotation of the question itself, which is shown in (15).

(15)

John-i

Bill-ul

ttayli-nya-ay-nta.

John-Nom

Bill-Acc

beat-Q-V-Decl

‘(Someone asks/They ask) whether John beat Bill.’

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 40

As introduced in Chapter 1, throughout this thesis I use prejacent as a purely descriptive term meaning a proposition without -ta-.

! 58

In (15), the entire question is embedded under a contracted quotative verb. (14), however, lacks this interpretation; the question is not embedded or quoted, but directly asked by the speaker.41 Now we can turn to the case of negation. With respect to sentential negation, -tanta also diverges from the quotative form. The only interpretation of negated sentences ending with -tanta is the negation of the prejacent. This is different from what we observe in negative sentences containing embedding quotative predicates, where the embedding quotative predicate is negated. This is shown in (16).

(16) a. Bill-un Bill-Top

John-i

na-lul po-ass-ta-ko

malha-cianh-ass-ta.

John-Nom

I-Acc

do-Neg-Past-Decl

see-Past-Decl-Comp

‘Bill did not say that John saw me’ b. John-i John-Nom

na-lul

po-cianh-ass-ta-nta.

I-Acc

see-Neg-Past-ta-Decl

‘John did not see me’ Implication: The speaker has reportative evidence that John saw the speaker.

This strongly suggests that -tanta in (16b) is not an elided form of an embedding quotative predicate, and therefore, -tanta is not semantically equivalent to an embedding quotative predicate. Notice indeed that the difference in (16) cannot be accounted for in terms of the position of negation. The examples above contain the so-called long form of Korean negation (!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 41

Barry Schein (p.c.) pointed out that the following analysis for -tanta could also be an alternative: in -tanta there is a covert quotatie verb, but this cannot be the target of a question simply because a question is always associated with a focus, but a phonologically null element cannot be focused. However, there are several other phenomena in Korean which can be accounted for by assuming that a phonological null element can be focused. For example, Guerzoni and Lim (2007) account for some semantic implications of concessive conditionals (or even if conditionals) in English and Korean by assuming that the focus associated with even is on a phonologically null functional head determining whether a given sentence is affirmative or negative. Similarly, Romero and Han (2004) accounts for the positive bias of preposed negation yes/no questions like Doesn’t John drink? by assuming an epistemic conversational operator VERUM, which is also phonologically null. Such phenomena can be regarded as evidence, if indirectly, that in Korean (and possibly in English) phonologically null elements can also be focused, and therefore can be a target of questioning.

! 59

cianh), which directly attaches after the predicate it negates. (17) shows that this negation cannot be inserted between -ta- and -nta, to negate the elided predicate in -tanta alone (if there is any).

(17)

??John-i

na-lul

po-ass-ta-cianh-nta.

John-Nom

I-Acc

see-Past-ta-Neg-Decl

Now compare (17) with (18).

(18) a. ?Bill-un Bill-Top

John-i

na-lul

po-ass-ta-ay-cianh-ass-ta.

John-Nom

I-Acc

see-Past-Decl-V-Neg-Past-Decl

John-i

na-lul

po-ass-nya-ay-cianh-ass-ta.

John-Nom

I-Acc

see-Past-Q-V-Neg-Past-Decl

‘Bill did not say that John saw me’ b. ?Bill-un Bill-Top

‘Bill did not say that John saw me’

In the examples in (18), which involve the phonological contraction of a quotative construction, long-form negation is shown to be much less degraded following the contraction of the embedding quotative predicate -ay- (corresponding to say in 18a and ask in 18b). Therefore, the distribution of long form negation constitutes another piece of evidence against identifying -tanta as a quotative predicate.

2.1.2.2. First-person effects

Following Aikhenvald (2004, section 7.2), I use the phrase ‘first-person effects of reportative evidentials,’ to refer to additional (non-evidential) implications of sentences with ! 60

reportative evidentials and first-person subjects, in which speaker describes their behaviors or feelings. Since speakers usually have direct evidence about their behaviors and feelings, it is unusual and counterintuitive for them to use reportative evidentials, unless they intend to emphasize the unusual nature of a situation where they lack direct evidence about their behaviors and feelings (e.g., because what they have heard is not true or because they are unconscious). Therefore, reportative evidentials (or non-firsthand evidentials in general) can only be used with the additional implication that the speaker does not recognize or is not aware of his/her behaviors or feelings or does not give much credence to what he/she is asserting. For example, in Jarawara (an Arauan language spoken in Amazonas, Brazil), if a speaker got drunk and is fully aware of it, the speaker may utter (19), where the firsthand evidential marker is used.

(19)

o-hano-hara

o-ke

1sg-be.DRUNK-IMM.P.FIRSTH.f 1sg-DECL.f ‘I got drunk (deliberately)’ (Aikhenvald 2004: 221, ex. 7.3)

However, if a speaker does not remember what happened before but heard from someone else that he/she got drunk, he/she may utter (20), where the non-firsthand evidential is used with the first person.

(20)

o-hano-hari

o-ke

1sg-be.DRUNK-IMM.P.NONFIRSTH.f

1sg-DECL.f

‘I got drunk’ Implication: the speaker does not recall it. (Aikhenvald 2004: 221, ex. 7.4) ! 61

The first-person effect of reportative evidentials can be used to test whether a given construction is a reportative evidential or a quotative construction. In the quotative construction in (21), where the subject of the embedded proposition is the first person, there is no first-person effect. Since the speaker of (21) just replicates Bill’s utterance that he/she got drunk yesterday, the speaker may not be aware that he/she got drunk yesterday or does not wish to make any commitment as to whether Bill’s utterance is true or not.

(21)

(Bill-un)

nay-ka ecey

chwiha-ess-ta-ko

malha-nta.

Bill-Top

I-Nom yesterday

drunk-Past-Decl-Comp

say-Decl

‘Bill says I got drunk yesterday’

However, with -tanta we get a first-person effect. (22), for example, can only be felicitously uttered when the speaker was not aware that he/she got drunk yesterday or doubts the information source (and thinks that he/she did not get drunk yesterday).

(22)

Nay-ka

ecey

chwiha-ess-ta-nta.

I-Nom

yesterday

drunk-Past-ta-Decl

‘I got drunk yesterday’ Implication: the speaker heard it from someone else and is not aware of it / doubts it.

One might think that this is due to the specific matrix subject, since in English impersonal quotation also seems to exhibit the same effect (Barry Schein p.c.). Consider the following two examples, where (23b), unlike (23a), shows the first-person effects, even though it is clearly a quotative construction. ! 62

(23) a. Bill says I got drunk yesterday. b. It is said I got drunk yesterday.

However, in Korean it seems that the quotative construction does not show any first person effects, no matter what there is an impersonal subject or not. Even though the subject in (21) is replaced by an impersonal subject, or is totally omitted, as in (24) (and therefore it is unclear who the source of this rumor might be), the sentence does not show any first-person effects:

(24)

(Salamtul-i)

Nay-ka ecey

(People-Nom) I-Nom yesterday

chwiha-ess-ta-ko

malha-nta.

drunk-Past-Decl-Comp

say-Decl

‘(People says that / it is said that) I got drunk yesterday’

Although I do not currently have an explicit analysis of first-person effects (and I do not know why there is difference between English and Korean), it is my impression that they are very much akin to Gricean conversational implicatures, as they emerge from an apparent diversion from cooperative communicative intent. What matters here is that, among the various forms of sentential endings and phonological contractions of embedding verbs of quotation that we are considering in this chapter, only -tanta contains a reportative evidential, since it is the only form that triggers a first-person effect.

2.1.3. Distributional diagnostics

As for distributional diagnostics, I propose two tests: ! 63

- Tense/aspect/modal morphemes - Embedding subjects

2.1.3.1. Tense/aspect/modal morphemes

It is well known that tense, aspect, and modal morphemes occupy separate functional heads in a structure.42 Given this, if in a given construction a quotative verb is contracted, prefinal endings such as tense/aspect/modal morphemes, which the main quotative predicate inflects with, may appear on the contracted form. This is exactly what happens in (25).

(25) a. John-un John-Top

Mary-ka

Bill-ul

manna-ess-nya-ay-ess-ta.

Mary-Nom

Bill-Acc

meet-Past-Q-V-Past-Decl

‘John questioned/asked of me whether Mary met Bill’ b. John-un John-Top

wuli-ka

Bill-ul

manna-ca-ay-ess-ta.

we-Nom

Bill-Acc

meet-Exh-V-Past-Decl

‘John suggested (to us) that we meet Mary’ c. John-un John-Top

nay-ka

Bill-ul

manna-la-ay-ess-ta.

I-Nom

Bill-Acc

meet-Imp-V-Past-Decl

‘John ordered (to me) that I meet Mary’

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 42

Authors disagree regarding which endings occupy independent functional heads. Choi K. (2003) claims that the honorific marker -si- is on Agr, and the past tense marker -ess- is on T, but the modal -keyss- and sentential endings such as -ta are in C. Yoon J.-M. (1990), on the other hand, argues that -keyss- is on Asp and distinguished from C. In this thesis, I assume that -keyss- also occupies a separate functional head. For the lexicalist approach (where all verbal inflections are generated in lexicon and none occupies a functional head position), see Sells (1995) and Yu and Sells (1995).

! 64

The examples in (26) are cases of phonological contraction of quotative constructions, where the contracted verb -ay- appears overtly. In these cases, the past tense morpheme can attach directly to the contracted verb -ay-. Similarly, in (25), the embedded clause is a simple declarative clause, and the contracted verb -ay- overtly appears with the past tense morpheme -ess-.

(26)

John-un

nay-ka

Bill-ul

manna-ess-ta-ay-ess-ta.

John-Top

I-Nom

Bill-Acc

meet-Past-Decl-V-Past-Decl

‘John told that I met Mary’

Turning now to -tanta, if -tanta is a free variant of -ta-ay-nta and, therefore, a quotative verb is elided in -tanta, we should expect tense/aspect/modal morphemes to surface between -ta and -nta. However, we observe that tense/aspect/modal morphemes do not surface with -tanta. If such morphemes are placed between -ta- and -nta, as in (27), the sentence is unacceptable.

(27) a. *Cwungtong-eyse Middle East-Loc

cencayng-i

ilena-ss-ta-ess-ta.43

war-Nom

occur-Past-Decl-Past-Decl

Intended: ‘(it is said that) a war broke out in the Middle East’ b. *Cwungtong-eyse Middle East-Loc

cencayng-i

ilena-ss-ta-koiss-ta.

war-Nom

occur-Past-Decl-Prog-Decl

Intended: ‘(it is being said that) a war broke out in the Middle East’ c. *Cwungtong-eyse Middle East-Loc

cencayng-i

ilena-ss-ta-lkes-i-ta.

war-Nom

occur-Past-Decl-Fut-Cop-Decl

Intended: ‘(it will be said that) a war broke out in the Middle East’ !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 43

One might think that (27a) is acceptable, but it is only acceptable when the pronunciation is the same as ta-ay-ess-ta, with phonological contraction of the embedding predicate.

! 65

d. *Cwungtong-eyse Middle East-Loc

cencayng-i

ilena-ss-ta-keyss-ta.

war-Nom

occur-Past-Decl-may-Decl

Intended: ‘(it may be said that) a war broke out in the Middle East’

Examples such as (27) show that -tanta cannot be accounted for in terms of phonological contraction of the quotative construction, because if a quotative verb were elided in -tanta, as in (27), pre-final endings should appear freely inside it, which is not the case. This supports our initial claim that -tanta is an instance of the reportative evidential construction in Korean and distinct from quotation.

2.1.3.2. Subjects for quotative predicates

In Korean, any argument can be relatively freely omitted, especially when it can be contextually recovered. Consider the following quotative construction:

(28)

(John-i)

Mary-ka

Bill-ul

manna-ess-ta-ko

malha-ess-ta.

John-Nom

Mary-Nom

Bill-Acc

meet-Past-Decl-Comp say-Past-Decl

‘John said Mary met Bill’

In uttering (28), one might elide the embedding subject, John-i, if John is a contextually salient speaker who said that Mary met Bill. Similarly, as another test to determine whether -tanta is a phonological contraction of the quotative construction, we may consider whether -tanta can appear with an embedding subject. Crucially, we find that -tanta is degraded with the main subject.

! 66

(29)

?*John-i John-Nom

cwungtong-eyse

cencayng-i

ilena-ess-ta-nta.

Middle East-Loc

war-Nom

occur-Past-ta-Decl

Again, if a quotative verb is elided in -tanta, we cannot account for the unacceptability of (29), where the subject of the allegedly elided embedding predicate appears overtly. Now compare (29) with (30), where the phonologically contracted embedding verb -ay- appears overtly.

(30)

?John-i

cwungtong-eyse

cencayng-i

ilena-ess-ta-ay-nta.

John-Nom

Middle East-Loc

war-Nom

occur-Past-Decl-V-Decl

‘John says a war broke out in the Middle East’

(30) is much more acceptable than (29), meaning that -tanta is actually a reportative evidential and not a free variant of a phonological contraction of the quotative construction. The claim that -tanta is actually a reportative evidential and incompatible with an overt embedding subject is further supported by cases of folklore, the first utterer of which is usually unclear. Hearsay evidentials only indicate that the source of information is someone other than the speaker but do not identify it. Quotative constructions, on the other hand, are used when there is a contextually salient utterer of the embedded proposition who the speaker is quoting. This leads to the prediction that only -tanta can be used with the folklore, which seems to be borne out, as shown in (31).

(31) a. *Yeysnalyeysnaley han wangkwuk-ey Once upon a time one kingdom-Loc

wangca han myeng-i

sal-ass-ta-ko

prince on CL-Nom

live-Past-Decl-Comp

malha-nta. say-Decl ! 67

b. Yeysnalyeysnaley han wangkwuk-ey Once upon a time one kingdom-Loc

wangca han myeng-i

sal-ass-ta-nta.

prince on CL-Nom

live-Past-ta-Decl

‘Once upon a time, there lived a prince in a kingdom’

(31) is a typical preface for a piece of folklore, the hero of which is a prince. Since the speaker of (31) does not know who the first utterer was (or who the source of information is), he/she cannot use the quotative form in (31a). (31b) seems much more acceptable. This also explains the fact that, unlike the quotation, -tanta is a reportative evidential.

2.1.4. Interim summary

Both the semantic diagnostics and the distributional diagnostics proposed above show that, contrary to the traditional analysis, -tanta (and -tapnita and -tay) cannot be reduced to phonological contractions of embedding predicates containing verbs of quotation. Given this, and given the type of implications that non-final -ta- introduces, I conclude that this particle is a genuine reportative evidential marker. Before concluding this section, I discuss two proposals that analyze other Korean morphemes as reportative (or hearsay) evidential markers.

2.1.5. Is the pro-verb form also a reportative evidential?

Kim N.-K. (1990) claims that pro-verb forms (examples repeated below), where the verb of saying is replaced by ha- ‘do’, are reportative evidential constructions and that -tanta is a phonological contraction of the pro-verb form -tako ha-nta: ! 68

(32) a. John-i John-Nom

Bill-ul

ttayli-ess-ta(-ko) ha-nta.

Bill-Acc

beat-Past-ta-Comp do-Decl

‘(Someone says/They say) John beat Bill’ b. John-i John-Nom

Bill-ul

ttayli-ess-ta(-ko) ha-pnita.

Bill-Acc

beat-Past-ta-Comp do-Decl

‘(Someone says/They say) John beat Bill’ Implication: the speaker heard that John beat Bill from someone else. c. John-i John-Nom

Bill-ul

ttayli-ess-ta(-ko) ha-e

Bill-Acc

beat-Past-ta-Comp do-Decl

‘(Someone says/They say) John beat Bill’

However, the pro-verb forms do not pass the reportative evidential diagnostics outlined above, and pro-verb forms behave more similarly to quotations than to reportative evidential constructions. Below, I use each of the diagnostics to test whether pro-verb forms are also reportative evidentials. First, consider the tense/aspect/modal test. If -ta-ko han-ta is a synchronically frozen construction whose function is to introduce reportative evidentiality, it should not show any inflection with tense/aspect/modal morphemes. Consider the following examples involving embedding predicates with the verb of saying.

(33) a. John-i John-Nom

cwungtong-eyse

cencayng-i

ilena-ss-ta-ko

Middle East-Loc

war-Nom

occur-Past-Decl-Comp

ha-ess-ta./malha-ess-ta. do-Past-Decl/say-Past-Decl ! 69

‘John said that a war broke out in the Middle East’ b. John-i John-Nom

cwungtong-eyse

cencayng-i

ilena-ss-ta-ko

Middle East-Loc

war-Nom

occur-Past-Decl-Comp

ha-koiss-ta./malha-koiss-ta. do-Prog-Decl/say-Past-Decl ‘John is saying that a war broke out in the Middle East’ c. John-i John-Nom

cwungtong-eyse

cencayng-i

ilena-ss-ta-ko

Middle East-Loc

war-Nom

occur-Past-Decl-Comp

ha-lkes-i-ta./malha-lkes-i-ta. do-Fut-Cop-Decl/say-Fut-Cop-Decl ‘John will say that a war broke out in the Middle East’

Now, compare the examples in (33) with those in (34), which Kim claims are unacceptable (from Kim N.-K. 1990: 426, ex. 15; the glosses and translations are mine).44

(34) a. Cwungtong-eyse Middle East-Loc

cencayng-i

ilena-ss-ta-ko

ha-ess-ta.

war-Nom

occur-Past-Decl-Comp do-Past-Decl

Intended: ‘(it is said that) a war broke out in the Middle East’ b. Cwungtong-eyse Middle East-Loc

cencayng-i

ilena-ss-ta-ko

ha-koiss-ta.

war-Nom

occur-Past-Decl-Comp do-Prog-Decl

Intended: ‘(it is being said that) a war broke out in the Middle East’

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 44

Kim N.-K. (1990) treats -keyss- and -lkes- as inferential modals like may, but we assume that -keyss- is an epistemic modal corresponding to may/must in English (degree of certainty may vary) and that -lkes- is a future morpheme.

! 70

c. Cwungtong-eyse Middle East-Loc

cencayng-i

ilena-ss-ta-ko

ha-lkes-i-ta.

war-Nom

occur-Past-Decl-Comp do-Fut-Cop-Decl

Intended: ‘(it will be said that) a war broke out in the Middle East’

According to Kim these examples are unacceptable, and the unacceptability (from Kim’s perspective) of the examples in (34) stems from the pro-verb form -tako hanta being a (syntactically frozen) reportative evidential construction. However, the problem is that, contrary Kim N.-K.’s (1990) intuition, the sentences in (34) are not as unacceptable as he claims. Kim himself is aware of this problem, and to get around it, he claims that -tako hanta is ambiguous between an embedding predicate where the verb of saying is replaced by the pro-verb (as in 32) and a frozen reportative evidential (as in 33). Kim further claims that the examples in (34) are acceptable only when the predicate does not involve a reportative evidential but a quotative predicate and the embedding subject is elided. This account of the acceptability of (34) is problematic. First of all, an ambiguity account of this sort should predict that when the sentence with the embedding verb of saying is acceptable, the sentence with the reportative evidential should also be acceptable. As a result, we do not have any precise way of distinguishing these two constructions, and Kim does not provide us with any semantic or syntactic criteria to distinguish the quotative meaning from the hearsay evidential meaning. We may adopt Aikhenvald’s (2004) distinction between hearsay and quotation, according to which hearsay evidentials are used when there is reported information without any reference to the source and quotations are used when the reported information has a salient source. In many cases, however, it is difficult to distinguish between information with or without a salient source. Furthermore, this distinction seems to contradict Kim’s proposal, as is shown by the next criterion.

! 71

Secondly, consider the test of the overt embedding subject. If a pro-verb form is derived from the ellipsis of a quotative verb, then an overt subject that the pro-verb is predicated of may appear with it. The following example shows that this prediction is actually borne out (from Kim N.-K. 1990, ex. 13a).

(35)

John-un

cwungtong-eyse

cencayng-i

ilena-ss-ta-ko

ha-nta.

John-Top

Middle East-Loc

war-Nom

occur-Past-Decl-Comp do-Decl

‘John says that a war broke out in the Middle East’

Kim N.-K. (1990) claims that (35) is unacceptable, but my judgment is that it is at least marginally acceptable. This marginality may be due to the tense of the main predicate, and not due to the hearsay evidential in the pro-verb form. If the speaker of (35) heard that a war broke out in the Middle East, (35) would be uttered later than the time when the speaker heard that a war broke out in the Middle East. However, the main predicate of (35) is in the present tense (regardless of whether the present tense is marked by a null morpheme Ø or -n- in -nta), and it is rather unnatural to quote someone else’s past utterance using the present tense. Given this, the prediction is that, if we changed the tense of the embedded and that of the embedding predicate accordingly, the sentence would be much more acceptable, and this prediction is borne out; (36) is more acceptable than (35).

(36)

John-un

cwungtong-eyse

cencayng-i

ilena-lswutoiss-ta-ko

John-Top

Middle East-Loc

war-Nom

occur-may-Decl-Comp

ha-lkes-i-ta. do-Fut-Cop-Decl ‘John will say that a war may break out in the Middle East’ ! 72

This said, we may conclude that Kim’s claim that the pro-verb form is a hearsay evidential is not supported by the (alleged) incompatibility of the main subject with the pro-verb form. Thirdly, the pro-verb form cannot be used when the source of the report is unclear, as in the case of folklore. Consider the previous example of folklore:

(37) a. *Yeysnalyeysnaley han wangkwuk-ey Once upon a time one kingdom-Loc

wangca han myeng-i

sal-ass-ta-ko

prince on CL-Nom

live-Past-Decl-Comp

malha-nta. say-Decl b. ??Yeysnalyeysnaley han wangkwuk-ey wangca han myeng-i Once upon a time one kingdom-Loc

sal-ass-ta-ko

prince on CL-Nom

live-Past-Decl-Comp

wangca han myeng-i

sal-ass-ta-nta.

prince on CL-Nom

live-Past-ta-Decl

ha-nta. do-Decl c. Yeysnalyeysnaley han wangkwuk-ey Once upon a time one kingdom-Loc

‘Once upon a time there lived a prince in a kingdom’

The quotation in (37a) is unacceptable as an introduction to folklore, because the speaker of (37a) does not know who first uttered this story. The pro-verb form in (37b) seems more acceptable than (37a), but it is still worse than (37c), where the reportative evidential -ta- is used. This provides further support for my argument that the pro-verb form is not a reportative evidential. Thus, the semantic diagnostics outlined above support my argument that the pro-verb form is not a reportative evidential. First, the pro-verb form does not exhibit first-person effects:

! 73

(38) a. (Bill-un)

nay-ka

ecey

chwiha-ess-ta-ko

ha-nta.

Bill-Top

I-Nom

yesterday

drunk-Past-Decl-Comp

do-Decl

‘Bill says that I got drunk yesterday.’ b. Nay-ka I-Nom

ecey

chwiha-ess-ta-nta.

yesterday

drunk-Past-ta-Decl

‘I got drunk yesterday’ Implication: The speaker heard it from someone and the speaker is not aware of it or doubts it.

Unlike the reportative evidential in (38b), the pro-verb form in (38a) does not introduce the implication that the speaker is unaware that he/she got drunk. Rather, it simply conveys what Bill said and does not carry any implication about the speaker’s awareness of his/her being drunk. Secondly, unlike -tanta, the pro-verb form can still be questioned or negated. First consider the test of question, the example of which is shown in (39):

(39)

Bill-un Tom-i

na-lul

po-ass-ta-ko

ha-ni?

Bill-Top Tom-Nom

I-Acc

drunk-Past-Decl-Comp

do-Q

‘Does Bill say that Tom saw me?’

(39) asks whether Bill says something or not, and not about whether Tom saw the speaker. However, the question in (40) asks whether Tom saw the speaker, and not whether there is any evidential implication.

(40)

Tom-i

na-lul

po-ass-ta-ni?

Tom-Nom

I-Acc

drunk-Past-ta-Q ! 74

‘Does Tom saw me?’ Implication: the speaker expects the addressee to answer whether Tom saw the speaker based on reportative evidence.

In addition to this, the pro-verb form can be negated, as shown in (41).

(41)

Bill-un

John-i

na-lul po-ass-ta-ko

ha-cianh-ass-ta.

Bill-Top

John-Nom

I-Acc

do-Neg-Past-Decl

see-Past-Decl-Comp

‘Bill did not say that John saw me’

In (41), what is negated is the fact that Bill said something, and not the fact that John saw the speaker. Thus, (41) is compatible with a situation in which John saw the speaker but Bill did not say anything about it. These two examples also show that the pro-verb form behaves in parallel with the quotation and should not be regarded as a reportative evidential. To summarize, contrary to Kim N.-K.’s (1990) claim, the pro-verb form is not a reportative evidential construction, but a variant of a quotation.

2.1.6. Self-quotation: is -ay- a hearsay evidential?

In the discussion above, I treated -ay as a phonological contraction of the pro-verb ha-. Chung J.-Y. (2009), however, claims that -ay- is a hearsay evidential, rather than a verbal predicate. Chung’s evidence comes from self-quotation, which Chung claims is possible in quotative constructions but not with hearsay evidentials. Consider the following example from Chung J. Y. (2009), which I have slightly revised: ! 75

(42)

(Scenario: I am having dinner at a restaurant with friends, including Tom and Mary. Tom tells Mary that he paid the bill, but Mary keeps asking whether the bill has been paid. When I tell someone else that Tom has already paid the bill, wondering why Mary keeps asking, I utter…) a. ton-ul Money-Acc b. ton-ul Money-Acc

nay-ess-tay-nun-tey

mwusun

iyaki-ha-nun-ke-ya?

pay-Past-Decl.hearsay-Top-conj what

say-do-Top-thing-Emph

nay-ess-ta(-ko) ha-nun-tey

iyaki-ha-nun-ke-ya?

mwusun

pay-Past-Decl-Comp do-Top-conj what

say-do-Top-thing-Emph

‘(Tom) said that he paid, what is Mary talking about?’ (Chung J.-Y. 2009, ex. 6a and 6b, respectively; glosses are Chung’s)

In (42a), -tay, which contains -ay, is used, whereas in (42b), the pro-verb (quotative) form is used. Suppose that, instead of Tom, I tell my friends that I have paid for the dinner and that Mary keeps asking who paid. According to Chung, I may utter (42b), but I cannot utter (42a). In other words, I can self-quote with the quotative form or the pro-verb form but not with the -tay form. This restriction is one reason Chung claims that -ay should be analyzed as a hearsay evidential, different from -tanta. Further evidence comes from the fact that -ay can appear with sentential types other than declaratives, such as interrogatives, imperatives, and exhortatives.

(43)

Interrogatives (Scenario: John comes back home and finds that his sister Jane is visiting. John asks Jane, “When did you come?”, but John’s voice is not loud enough, and Jane does not hear him. Mary, John’s wife, repeats John’s question to Jane, saying…) ! 76

(John-i)

ney-ka

encey

o-ass-ni-ay.

(John-Nom)

You-Nom

When

come-Past-Q-ay

‘(John asked) when you came.’

(44)

Imperatives (John-i)

hakkyo-ey

ka-la-ay.

John-Nom

school-Loc

go-Imp-ay

‘(John ordered me/us to) go to school.’

(45)

Exhortatives (John-i)

hakkyo-ey

ka-ca-ay.

John-Nom

school-Loc

go-Exh-ay

‘(John suggested to us that we) go to school.’

According to Chung J.-Y. (2009), examples such as (43) through (45) show that -ay takes wide scope over questions, imperatives, and exhortatives, because the evidential implication introduced by -ay is illocutionary and not truth-conditional. By using -ay, the speaker presents what someone else said and does not make any commitment to whether the prejacent is true. However, Chung’s evidence strikes me as inconclusive. First, the judgments are not clear. My intuition is that (42a) is much better if the subject of the prejacent nay ‘I’ appears overtly, as in (42'):

(42')

?Nay-ka ton-ul I-Nom

nay-es-tay-nun-tey

mwusun iyaki-ha-nun-ke-ya?

money-Acc pay-Past-Decl.hearsay-Top-conj what

say-do-Top-thing-Emph

‘I said that I paid. What are you talking about?’ ! 77

The other reason I do not endorse Chung’s claim is that, as we saw above, tense/aspect/modal morphemes take wide scope over -ay. Consider the following examples:

(43')

(John-i)

ney-ka

encey

o-ass-nya-ass-e. (=>o-ass-ny-ay-ess-e)

(John-Nom)

You-Nom

When

come-Past-Q-ass-Decl

‘(John asked) when you came.’

(44')

(John-i)

hakkyo-ey

ka-la-ay-ess-e.

John-Nom

school-Loc

go-Imp-ay-Past-Decl

‘(John ordered me/us to) go to school.’

(45')

(John-i)

hakkyo-ey

ka-ca-ay-ess-e

John-Nom

school-Loc

go-Exh-ay-Past-Decl

‘(John suggested to us to) go to school.’

In these examples, the past tense marker -ess seems to take wide scope over -ay. For example, in (43'), the speaker quotes John’s question, and John’s uttering the question precedes the utterance of (43'). The speaker of (43') indicates this relationship by using the past tense and in this sense, the past tense in (43') takes wide scope over the quotation of the question. Similarly, in (44'), the past tense takes wide scope over the imperative, and in (45'), the past tense takes wide scope over the exhortative. This is unexpected in Chung’s analysis. If -ay were an illocutionary operator, as Chung J.-Y. (2009) claims, it would always take wide scope other sentential operators, including negation, modals, and tense, contrary to the facts.

! 78

To conclude, Chung’s claim that -ay is a hearsay evidential is empirically incorrect. Examples such as those in (43') through (45') suggest that -ay is more likely a phonological contraction involving a quotative predicate. Having established this much, we need to return to the paradigm of Korean final endings and address one more issue. The paradigm is repeated below.

(46)

Simple

Complex

Declarative

Question

Declarative

Formal Style

-pnita

-pnikka

-ta-pnita

Plain Style

-(n)ta

-ni

-ta-nta

-e/a

-ta-e/a (=> tay) -ta-e/a (=> tay)

Intimate Style -e/a

Question -ta-pnikka -ta-ni

The last question that needs to be addressed is whether -tay is also a phonological contraction of the quotative construction. My view is that -tay in declaratives is actually ambiguous. It can either be derived from the declarative ending -ta followed by -ay-, from the phonological contraction of the pro-verb followed by the declarative ending -e, or from the reportative evidential -ta- followed by the declarative ending -e. For brevity, I use only two of the various diagnostics: the past tense morpheme test and the folklore case. First consider the diagnostics of the past tense morpheme. The prediction is that when -tay involves the reportative evidential -ta-, the past tense morpheme attaches to the main predicate directly and, therefore, will appear before -ta-, but when -tay involves the phonological contraction of the embedding quotative verb, the past tense morpheme will appear after -tay. This prediction is borne out.

(47) a. John-i John-Nom

Mary-lul

ttayli-ess-ta-e

(=>ttayli-ess-tay)

Mary-Acc

beat-Past-ta-Decl ! 79

‘John beat Mary’ Implication: the speaker has heard from someone else that John beat Mary. b. (Tom-un) Tom-Top

John-i

Mary-lul

ttayli-ess-ta-ay-ess-e

John-Nom

Mary-Acc

beat-Past-Decl-V-Past-Decl

‘(Tom said) John beat Mary’

The difference between -tay from -ta- and a declarative and -tay from the embedding quotative predicate can be further shown by testing whether the past tense morpheme can appear after -tay in the cases of folklore, where only the reportative evidential is used. The prediction is that -tay followed by the past tense marker should not appear in cases of folklore, since the past tense marker can attach after -tay only when -tay contains an embedding verb of quotation, a construction which is not available in cases of folklore. This prediction is also borne out:

(48) a. Yeysnalyeysnaley han wangkwuk-ey Once upon a time one kingdom-Loc

wangca han myeng-i

sal-ass-ta-e.

prince on CL-Nom

live-Past-ta-Decl

b. *Yeysnalyeysnaley han wangkwuk-ey wangca han myeng-i sal-ass-ta-ay-ess-e. Once upon a time one kingdom-Loc

prince on CL-Nom

live-Past-Decl-V-Past-Decl

‘Once upon a time, there lived a prince in a kingdom’

To summarize, these two tests show that -tay is actually ambiguous between a reportative evidential with a declarative marker and the phonological contraction of an embedding predicate involving a verb of saying.45

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 45

However, there are some examples where -tay behaves differently from -tanta and -tapnita. I return to this issue in Chapter 5, particularly when discussing the ‘affectionate’ or ‘intimacy’ reading of Korean reportative evidentials.

! 80

2.1.7. Interim summary: non-final -ta- as reportative evidential

In this subsection, I argued that Korean has reportative evidential constructions that are distinct from quotative constructions. Among the three forms—quotative, pro-verb, and -tanta—only tanta is a reportative evidential, the other two constructions being quotations.46 I presented the following evidence in support of this claim:

"

Unlike in the quotative form or pro-verb form, implications introduced by -tanta cannot be negated or questioned.

"

Only -tanta shows the first-person effect.

"

Additional tense/aspect/modal morphemes cannot appear within -tanta.

"

The main subject can appear with the quotative form and pro-verb form, but not with -tanta.

I also pointed out that Chung J.-Y.’s (2009) claim that Korean -ay is a hearsay evidential is not empirically tenable. My conclusions here are novel. While Kim N.-K. (1990) proposes that the pro-verb form -tako ha- is a reportative evidential, and Kim N.-K. (2001) and Chung K.-S. (2005) propose that tanta is a reportative evidential, these analysis do not account for why -tako ha- is similar to an embedding predicate involving a verb of saying nor why -tay and -tapnita also introduce the reportative evidential implication. My proposal is that the non-final -ta- is the only reportative evidential morpheme in Korean (Lim 2009), a conclusion based on the morphological paradigm of Korean sentential endings that I reviewed at the beginning of this chapter.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 46

Furthermore, as we will see in Chapter 3, -tanta introduces non-truth-conditional meaning, whereas the other two forms make contributions to the truth-conditions of the sentences in which they appear.

! 81

2.2. Direct evidentiality in Korean

Sohn (1994) claims that direct evidentiality in Korean is introduced by the morpheme -te-. For example, in (49), -te- introduces the implication that the speaker has direct (perceptive) evidence that John met Mary.

(49)

John-i

ecey

Mary-lul

manna-te-la.47

John-Nom

yesterday

Mary-Acc

meet-te-Decl

‘John met Mary yesterday’ Implication: the speaker has direct evidence (e.g., the speaker saw) that John met Mary.

However, an analysis of -te- as expressing direct evidentiality appears at first problematic. First, without any overt tense marker, sentences using -te- describe events in the past rather than the present. In (49), the time adverb ‘yesterday’ indicates that John’s meeting Mary occurred in the past. Why does a sentence with -te- express eventuality in the past without any past tense morpheme? This has led some authors to analyze -te- as a special kind of past tense marker (e.g., Chung K.-S. 2005, 2007). The second - and perhaps more crucial - problem with analyzing -te- as an evidential morpheme is that when -te- appears with other tense/aspect/mood morphemes, the evidential implication it introduces appears to be inferential. Consider the following examples:

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 47

-la is a declarative ending that appears with -te-.

! 82

(50) a. (Scenario: yesterday I saw John meet Mary) John-i

ecey

Mary-lul

manna-te-la.

John

yesterday

Mary-Acc

meet-te-Decl

‘John met Mary yesterday’ Implication: the speaker has direct evidence that John met Mary. b. (Scenario: yesterday I found that John had Mary’s book, which I saw before in Mary’s home) John-i

kucey

Mary-lul

manna-ess-te-la.

John-Nom

the-day-before-yesterday

Mary-Acc

meet-Past-te-Decl

‘John met Mary the day before yesterday’ Implication: based on the speaker’s inference, John met Mary the day before yesterday. c. (Scenario: John has some problem with his homework and needs help as soon as possible. I think Mary is the only one who can help him, but she is out of town today and will be back tomorrow) John-i

nayil-i-lato

Mary-lul

manna-keyss-te-la.48

John-Nom

tomorrow-L-at least/even

Mary-Acc

meet-may-te-Decl

‘John will meet Mary tomorrow at the latest (epistemic)’ Implication: based on the speaker’s inference, John will meet Mary tomorrow.

In (50b) and (50c), where overt tense/modal elements appear in the prejacent, the sentence carries the implication that the speaker’s assertion is based on inference from past perceptual evidence. This shift from direct to inferential evidentiality needs to be accounted for.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 48

As already mentioned above, throughout this chapter, I gloss -keyss- as may, simply indicating that this element has (epistemic) modal force. Some authors treat-keyss- as a ‘future-oriented modal element’ (e.g., Lee J. 2010 and references therein), but others analyze it as a future tense marker (e.g., Ko Y. 2007). I adopt the former view, but the exact nature of -keyss- is irrelevant at the moment.

! 83

Based on these observations, Chung K.-S. (2005, 2007) proposes that -te- is a specific type of past tense marker, which she dubs deictic past tense and which Faller (2004) defines as a tense marker referring to a specific time as well as a location. Here is a very rough summary of Chung’s (2005, 2007) argument. Given examples such as (50b) and (50c), -te- does not seem to be a direct evidential. Examples such as (50a) show that when there is no overt tense marker, a sentence with -te- always describes an event that occurred before the time of utterance. This leads Chung to conclude that -te- is a past tense marker. It is not a simple past tense marker, however, but a deictic past tense marker, in that it not only encodes information about a specific time point, but about a specific spatial point. Consider the examples in (51), which show that the temporal adverbial cikum ‘now’ is compatible with -te-.

(51)

cikum

pakk-ey-nun

pi-ka

o-te-la.

Now

outside-Loc-Top

rain-Nom

come-te-Decl

‘[I noticed] It is raining outside now.’ (Chung K.-S. 2007: 21)

Now, compare (51) with (52).

(52)

#cikum

yeki-eyse-nun pi-ka

o-te-la.

Now

here-Loc-Top

come-te-Decl

rain-Nom

(Intended) ‘[I noticed] It is raining here now.’

According to Chung, the acceptability of (51) (and the unacceptability of 52) is accounted for by assuming that -te- is a deictic past tense marker. In other words, roughly speaking, that -te“denotes a certain past time when the speaker either perceived a given event itself or some ! 84

indirect evidence of the event” (Chung K.-S. 2007: 195). More precisely, a sentence with -te- is true if and only if, at a certain time before the time of utterance, the speaker directly or indirectly perceived an event in a place other than the place of utterance. If -te- is a deictic past tense marker, where do we get the evidentiality in sentences with te-? Chung proposes that -ess- and -keyss-, which are traditionally analyzed as a past tense marker and a future (or modal) marker, respectively, are ambiguous between tense markers and evidential markers. For the present tense marker, Chung assumes a null morpheme, Ø, which is also ambiguous between a present tense marker and an evidential marker. Specifically, when there is no deictic tense marker, these morphemes are assumed to behave as non-evidential tense/aspect/modal markers, as in (53), but when there is a deictic tense marker, such as -te-, they behave as evidential markers, as in (54) (from Chung K.-S. 2007: 205):

(53)

(54)

(when there is no deictic tense) Ø:

present tense

-ess-:

perfective

-keyss-:

mood

(when there is an overt deictic tense) Ø:

direct evidential

-ess-:

indirect evidential based on result states

-keyss-:

indirect evidential based on reasoning

Under this analysis, the previous examples would be treated as follows: In (49) (and therefore in 50a), since there is a -te-, the morpheme Ø behaves as a direct evidential marker, introducing the implication that the speaker directly perceived (or saw) that John met Mary. In (50b), with the ! 85

deictic past tense marker -te-, -ess- behaves as an indirect evidential marker. Therefore, in uttering (50b), the speaker has indirect evidence about the fact that John met Mary, namely, that John has Mary’s book. Similarly, in uttering (50c), the speaker has indirect evidence about the fact that John will meet Mary, namely, reasoning based on Mary’s schedule and John’s situation. In (51), the adverbial phrase pakk-ey ‘outside’ indicates that the speaker perceived the event of raining in a place different from the place of utterance. Even though there is a time adverbial cikum in (51), the perception of raining outside is still before the time of utterance. There are several problems with Chung’s argument, however. For detailed criticisms of Chung K.-S. (2005, 2007), I refer readers to Lee J. (2010) and in this subsection suffice it to point out some problems in Chung’s analysis. First, the analysis needs to assume unwanted ambiguity in the tense/modal morphemes -ess- and -keyss-. Chung claims that -ess- and -keyss- are tense/modal elements in sentences without deictic tense makers and evidentials in sentences with deictic tense markers, but there is no explanation for why -ess- and -keyss- do not have evidential meaning without -te-. Second, Chung’s analysis overlooks evidence that the behavior of -teparallels that of epistemic modals (see, e.g., Lee J. 2010, some of whose evidence I review in the next chapter). Finally, and more crucially, there is no need to introduce the deictic tense in Korean in order to account for the data presented above. For example, Lee J. (2010) proposes the following lexical entry for -te-, which seems compatible with the interpretations illustrated above:

(55)

[[-te-]] = %P. %w. %t. &w’ [(w’'BEST(S0, ST/DX, w, t) " t)now ) * P(w’)(t)] (where BEST(S0, ST/DX, w, t) is a function from world-time pairs to sets of worlds that are ranked according to the ordering source ST/DX(w,t) among the worlds determined by the modal base S0(w,t), thus, the worlds in BEST(S0, ST/DX, w, t) are those that are compatible with what the speaker knows based on sensory observations in w at t and are

! 86

ranked according to his/her expectations about what the world w was/is/would be like prior to/at/after t). (Lee J. 2010: 10)

I discuss the semantics of -te- in more detail in the following chapters. I also show that (55) is not fully satisfactory to account for the semantics of -te-, but for now, suffice it to say that there are several alternatives available to account for the different implications apparently introduced by te- in (55), without appealing to the notion of deictic tense. To conclude, Chung’s (2005, 2007) claim that -te- is a deictic past tense marker is not convincing. Therefore, following several authors, including Sohn H. (1994), Cinque (1999: adopting Sohn H. 1994), Song J. (1998), and Lee J. (2010), I treat -te- as a direct evidential marker.

2.3. Inferential evidentiality in Korean

At first glance, there is no counterpart of inferential evidentiality to direct evidentiality te- or reportative evidentiality -ta- in Korean. However, Kwon I. (2010) reports that several periphrastic (and rather grammaticalized) forms are employed in Korean to express inferential evidentiality. For example, -napo and nmoyangi- in (56) (from Kwon I. 2010) indicate that the speaker’s assertion is based on an inference from direct perception.

(56)

(Scenario: entering John’s room in the morning, the speaker finds that John is not there) John-i

hakkyo-ey

ka-ess-napo-a/ka-nmoyang-i-a.

John-Nom

school-Loc

go-Past-napo-Decl/go-nmoyang-Cop-Decl ! 87

‘John went to school’ Implication: the speaker’s assertion is based on an inference. (Slightly modified from Kwon I. 2010, ex. 2)

-napo- originates from the verbal ending -na- ‘whether’ and the verb po- ‘see’, and -nmoyangoriginates from the relative marker -n-49 and the noun moyang ‘form/shape.’ Without -na- or -n-, respectively, the verb po- and the noun moyang preserve their lexical meanings, as shown in (57).

(57) a. Na-nun I-Top

ecey

John-ul

po-ass-ta.

yesterday

John-Acc

see-Past-Decl

‘I saw John yesterday’ b. I sangca-uy this box-Gen

moyang-i

isangha-ta.

shape-Nom

weird-Decl

‘The shape of this box is weird’

Kwon I. (2010) shows that -napo- and -nmoyang- in (55) have characteristics of inferential evidentiality (unlike in 56) and tries to account for their various usages in terms of inferential evidentiality. The diagnostics presented above yield different results, however. Some seem to support Kwon’s claim, but others seem to disconfirm it. Among the various diagnostics, I apply three here for brevity: the negation test, the tense marker test, and the question test. The negation test supports Kwon’s claim. The implications introduced by -na- and -moyangi- cannot be negated. Consider (58), where the negation an is inserted.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 49

Kwon I. (2010) glossed this -n- as perfective marker.

! 88

(58)

(Scenario: entering John’s room in the morning, the speaker finds that John is not there) John-i

hakkyo-ey

an

ka-ess-napo-a/ka-nmoyang-i-a.

John-Nom

school-Loc

Neg

go-Past-napo-Decl/go-nmoyang-Cop-Decl

‘John did not go to school’ Implication: the speaker’s assertion is based on inference.

The negation an cannot negate the evidential implication introduced by -na-/-nmoyang-. Furthermore, using the long-form negation after -na-/-nmoyang- simply results in an unacceptable sentence.

(59)

*John-i

hakkyo-ey

ka-ess-napo-cianh-a/ka-nmoyang-i-cianh-a.

John-Nom

school-Loc

go-Past-napo-Neg-Decl/go-nmoyang-Cop-Neg-Decl

However, the tense marker test yields mixed results. For example, -napo- cannot be followed by a past tense marker, whereas -nmoyangi- can.

(60)

(Scenario: entering John’s room in the morning, the speaker finds that John is not there) a. *John-i John-Nom b. John-i John-Nom

hakkyo-ey

ka-napo-ess-ta.

school-Loc

go-napo-Past-Decl

hakkyo-ey

ka-nmoyang-i-ess-ta.

school-Loc

go-nmoyang-Cop-Past-Decl

‘John went to school’ Implication: the speaker’s assertion is based on inference.

! 89

As for the question test, in general napo/nmoyang do not appear in questions. When they appear in a question, it becomes simply unacceptable for Korean native speakers, regardless of whether it is a yes/no question or a wh-question:

(61)

(62)

*John-i

hakkyo-ey

ka-ess-napo-ni/ka-nmoyang-i-ni?

John-Nom

school-Loc

go-Past-napo-Q/go-nmoyang-Cop-Q

*Nwu-ka

hakkyo-ey

ka-ess-napo-ni/ka-nmoyang-i-ni?

Who-Nom

school-Loc

go-Past-napo-Q/go-nmoyang-Cop-Q

I do not know at the moment why it is so, but at least the question test does not help us to decide whether napo/nmoyang is an evidential morpheme or not. Since the tests are inconclusive, I do not provide a full formal analysis of these morphemes here, but simply point out that as Kwon (2010) proposed, it is possible that in Korean, synchronically speaking, these morphemes are inferential evidential markers.

2.4. Summary and conclusion

In this chapter, I outlined various types of evidentiality in Korean. I focused primarily on reportative evidentiality, proposing that we need to posit a reportative evidential in Korean. I also argued against Chung J.-Y. (2009), claiming that we do not need to posit a hearsay evidential that can take non-declarative sentential types (such as interrogatives, imperatives, and exhortatives), but rather that what look like hearsay evidentials with non-declarative sentential types are actually quotative constructions. Furthermore, unlike Chung K.-S. (2005, 2007), I followed Sohn ! 90

H.-M. (1994), Song (2002), and Lee J. (2010) in assuming that the Korean pre-final verbal ending -te- expresses direct evidentiality. Finally, I briefly reviewed Kwon I.’s (2010) discussion of inferential evidentiality in Korean, pointing out that such constructions can also be analyzed as inferential evidentials, though more investigation is required. In summary, in this thesis, I assume that in Korean at least two evidential morphemes, direct -te- and reportative -ta- exist, and I will mainly focus on these two evidentials. I also agree that there may also be inferential evidentials in Korean, but due to the inconclusive results of the tests, I will not discuss these markers.

(63)

Korean evidentials a. Direct evidential: -teb. Reportative evidential: -ta(c. Inferential evidential: -na-, -nmoyang-)

In the next chapter, I analyze the semantic nature and linguistic category of the Korean evidential markers, showing that they are similar to Izvorski’s (1997) Perfective Evidentials and Matthewson et al.’s (2007) St’át’imcets evidentials in that they introduce evidential presuppositions rather than illocutionary meanings. However, I will also point out that Korean evidentials do not exactly pattern with PEs or St’át’imcets evidentials, in that they do not behave like epistemic modals.

! 91

Chapter 3 Semantics of Korean Evidentiality

In this chapter, I investigate the semantic nature of Korean evidentiality. As mentioned in the introduction, establishing the semantics and pragmatics of evidentials involves addressing two main questions. The first concerns the content of evidentiality, specifically, whether we find components of epistemic modality or of illocutionary force in evidentials in a language. The second question concerns the level of interpretation (in the broad sense) on which they operate, e.g., on the level of presuppositions or of illocutionary acts, etc. We have also seen that there are two approaches to the nature of the narrow evidentiality, which diverge depending on the answers to those two questions. The first treats evidentials as containing epistemic modality components and treats evidential implications as presuppositions (e.g., Izvorski 1997, Mathewson et al. 2007). The second treats evidentials as illocutionary operators, introducing evidentiality as illocutionary forces (e.g., Faller 2002). Many authors agree that evidentials behave differently with respect to these questions in different languages, which means that the semantic nature of the narrow evidentiality may differ across languages (although it is not theoretically desirable to apply different theories to different evidentials in different languages). Thus, it is possible that in one language evidentials are epistemic modals introducing evidentiality as presuppositions, and in another, they are illocutionary operators, which introduce evidentiality as illocutionary forces. In this chapter, I first argue that evidentiality in Korean does not make any contribution to truth-conditions and, specifically, that it operates on the level of presuppositions. That is, Korean evidentials introduce definedness conditions regarding the source of the speaker’s information. In addition to this, I argue that there is no convincing argument to assume that Korean evidentials ! 92

contain components of epistemic modality. To support these claims, I apply various diagnostics proposed in Faller (2002) and Matthewson et al. (2007) to the Korean evidentials -ta- and -te-. As in the previous chapter, my discussion mainly focuses on the reportative evidential -ta-, and I briefly discuss -te- in each section.50 Finally, in the last section, I will show one interesting aspect of the Korean reportative -ta-, that is, the apparent variation of the degree of certainty the speaker has when using -ta-, and review Matthewson et al. (2007), where variability of the degree of certainty in St’át’imcets evidentials is reported and analyzed.

3.1. Korean evidential implication as presupposition

In this section, I apply several criteria that establish whether the implications introduced by Korean evidentials are either truth-conditional, presuppositional, or illocutionary. I argue that the evidentiality introduced by -ta- does not make any contribution to the truth-conditions of the hosting sentence. Among the various levels of meaning that are not truth-conditional, including implicatures, presuppositions, and illocutionary forces, I claim that the evidentiality of -ta- is a presupposition.

3.1.1. Question and negation

Let us begin with a rather standard test, namely, questions and negation. We saw in the previous chapter that if a semantic object contributes to the truth-conditions of a proposition, then it should be possible to question or negate it, whereas if it is not truth-conditional, it cannot be questioned or negated. Note that these two tests only distinguish elements that contribute to truth!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 50

Throughout this chapter, the evidential implication is either indicated separately, as before, or indicated using square brackets.

! 93

conditions from those that do not; they do not distinguish whether elements operate on the level of presuppositions or illocutionary forces. However, note further that the presupposition trigger and the illocutionary operator are not subject to negation or question for different reasons. Presuppositions are projected under question or negation, whereas illocutionary operators are on an entirely different linguistic plain, applying only after all other linguistic elements are composed. With this in mind, consider (1), where -ta- is used in a question.

(1)

John-un

Mary-lul

po-ass-ta-ni?

John-Top

Mary-Acc

see-Past-ta-Q

‘Did John see Mary?’ Implication: the addressee has reportative evidence about whether John saw Mary. NOT: ‘Does John have reportative evidence that John saw Mary?’

What is asked in (1) is not whether the addressee has heard (or has reportative evidence) about whether John saw Mary. The speaker only expects the addressee to answer based on reportative evidence, but the question is about whether John saw Mary. Based on this, it suffices to conclude that at least as far as this test is concerned, the evidentiality introduced by -ta- is not truthconditional. (2) likewise shows that the reportative evidential meaning of non-final -ta- is not truthconditional. If the implication of non-final -ta- were truth-conditional, it should be possible to negate it. However, (2b) does not mean: ‘John beat his own dog, but the speaker did not have reportative evidence.’ Rather, the prejacent is negated, and the evidential implication is not.

! 94

(2) a. John-i John-Nom

caki kay-lul

ttayli-ess-ta-nta.

self dog-Acc

beat-Past-Evi-Decl

‘John beat his own dog.’ Implication: the speaker has reportative evidence that John beat his own dog. b.

John-i

caki kay-lul

an

ttayli-ess-ta-nta.

John-Nom

self dog-Acc

Neg

beat-Past-Evi-Decl

‘John did not beat his own dog.’ Implication: the speaker has reportative evidence that John did not beat his own dog. Absent implication: the speaker’s evidence that John beat his own dog is not reportative.

The same argument can be applied to -ta- in questions. (3b), where -ta- is used in a question containing negation, does not mean, ‘Did you not have any reportative evidence that John beat his own dog?’ but rather ‘Did John not beat his own dog?’ with the implication that the addressee is expected to answer based on reportative evidence.

(3) a. John-un John-Top

caki kay-lul

ttayli-ess-ta-ni?

self dog-Acc

beat-Past-Evi-Q

‘Did John beat his own dog?’ Implication: the addressee is expected to answer based on reportative evidence. b. John-un John-Top

caki kay-lul

an

ttayli-ess-ta-ni?

self dog-Acc

Neg

beat-Past-Evi-Q

‘Did John not beat his own dog?’ Implication: the addressee is expected to answer based on reportative evidence. Absent implication: the speaker is asking whether the addressee has reportative evidence about whether John beat his own dog. ! 95

Both examples indicate that the reportative evidential implication cannot be negated and, therefore, is not truth-conditional. As with -ta-, the implication introduced by -te- cannot be negated or questioned, which suggests that it is not truth-conditional either.

(4)

John-i

Mary-lul

an

manna-te-la.

John-Nom

Mary-Acc

not

meet-te-Decl

‘John did not meet Mary’ Absent reading: the speaker did not see that John met Mary. Actual implication: the speaker saw that John did not meet Mary.

(5)

John-i

Mary-lul

manna-te-nya?

John-Nom

Mary-Acc

meet-te-Q

‘Did John meet Mary?’ Absent reading: Do I/you have direct evidence that John met Mary? Actual implication: the addressee is expected to answer based on direct evidence.

These two examples show that evidential implications introduced by -te- are not truth-conditional. However, as with -ta- it remains unclear whether -te- is a presupposition trigger or an illocutionary operator.

! 96

3.1.2. Truth/falsity of the assertion

As was pointed out in the introduction of this chapter, both the presuppositional and illocutionary views on evidentiality converge relative to the previous two tests. The negation and question tests show only that the evidential implication introduced by an evidential marker is not truth-conditional; they do not show whether it is a presupposition or an illocutionary force. How then can we distinguish between these two views? Faller (2002) suggests that they make different predictions when a sentence containing the relevant markers is followed by a sentence showing that the prejacent is true or that it is false. Illocutionary operators of the sort that Faller (2002) identifies with a reportative evidential in Cuzco Quechua are such that the prejacent is simply presented without any commitment by the speaker as to the truth or the falsity of the prejacent. Given this, such an operator should always be felicitous in a context where the speaker knows whether the prejacent is true or false. According to Faller (2002), this is the case with the Cuzco Quechua reportative evidential -si-, as shown in (6).

(6)

para-sha-n-si,

ichaqa

mana

crei-ni-chu

rain-PROG-3-si

but

not

believe-1-NEG

‘It is raining, but I don’t believe it.’ EV=

speaker is/was told that it is raining. (Faller 2002: 194)

However, if an evidential marker is merely a presuppositional operator, a sentence with that evidential marker is infelicitous, because the assertion of a sentence with an evidential presupposition trigger ultimately amounts to the prejacent itself. Therefore, if the prejacent is ! 97

presupposed to be false, the sentence is infelicitous due to contradiction, and if the prejacent is presupposed to be true, it is infelicitous due to redundancy (a speaker does not need to repeat what is presupposed to be true: for a detailed discussion, see Matthewson et al. 2007, sections 4.1 and 4.2). The following example shows that the reportative evidential -ta- is a presupposition trigger and that its implication is not illocutionary. (7) shows that if the speaker is already sure of the truth or falsity of the proposition, non-final -ta- cannot be used felicitously.

(7)

(Context: I did some work for a company, and they said they put my pay, $200, into my bank account. However, they actually did not pay me at all.) (adapted from Matthewson et al. 2007: 214) #Hoysa-ka

na-hanthey 200 talle-lul

company-Nom I-Dat Haciman hoysa-nun

cwu-ess-ta-nta.

200 dollar-Acc give-Past-ta-Decl 200 talle-lul

na-hanthey cwu-cianh-ass-e.

However company-Top 200 dollar-Acc I-Dat

give-Neg-Past-Decl

‘(Reportedly) the company gave me 200 dollars, but they did not give me 200 dollars.’

(8)

(Context: I did some work for a company, and they said they put my pay, $200, into my bank account. I checked the account and saw that $200 had been deposited.) #Hoysa-ka

na-hanthey 200 talle-lul

company-Nom I-Dat

cwu-ess-ta-nta.

200 dollar-Acc give-Past-ta-Decl

Kuliko hoysa-nun

200 talle-lul

na-hanthey cwu-ess-e.

And company-Top

200 dollar-Acc

I-Dat

give-Past-Decl

‘(Reportedly) the company gave me 200 dollars, and they gave me 200 dollars.’

! 98

Since the assertion with -ta- is not felicitous when the prejacent is known to be true or false, as in (8), we can conclude that the reportative evidential implication introduced by -ta- is not illocutionary but presuppositional. Like -ta-, -te- cannot be used felicitously when the truth or falsity of the prejacent is known, which shows that the implication introduced by -te- is also presuppositional.

(9)

(Context: I did some work for a company, and they said they put my pay, $200, into my bank account. I checked my account and saw that $200 had not been deposited.) #Hoysa-ka

na-hanthey

200 talle-lul

cwu-ess-te-la.

company-Nom I-Dat

200 dollar-Acc give-Past-te-Decl

Haciman

hoysa-nun

200 talle-lul

However

company-Top 200 dollar-Acc I-Dat

na-hanthey

cwu-cianh-ass-e. give-Neg-Past-Decl

‘(I saw) the company gave me 200 dollars, but they did not give me 200 dollars.’

(10)

(Context: I did some work for a company, and they said they put my pay, $200, into my bank account. I checked the account and saw that $200 had been deposited.) #Hoysa-ka

na-hanthey 200 talle-lul

company-Nom I-Dat Kuliko hoysa-nun And

cwu-ess-te-la.

200 dollar-Acc give-Past-te-Decl 200 talle-lul

company-Top 200 dollar-Acc

na-hanthey cwu-ess-e. I-Dat

give-Past-Decl

‘(I infer) the company gave me 200 dollars, and they gave me 200 dollars.’

Before moving on to the next section, let me point out that this pattern also brings to mind a puzzle raised about epistemic modals (see, e.g., Yalcin 2007), according to which a sentence with

! 99

a possibility epistemic modal cannot be uttered felicitously when the truth or falsity of the prejacent is known, as shown in (11a-d) (from Yalcin 2007: 983, ex. 1-4):

(11) a. #It is raining and it might not be raining. b. #It is raining and possibly it is not raining. c. #It is not raining and it might be raining. d. #It is not raining and possibly it is raining.

It is beyond the scope of this paper to show why epistemic modals pattern as in (11) (see Yalcin 2007 for a proposal), but such examples suggest that there is some parallelism between evidentials and epistemic modals, even though it is not clear whether evidentials can also be accounted for in terms of epistemic modals or not. In the next section I return to the question of whether Korean evidentials are epistemic modals or not.

3.1.3. Accommodation and correction

Presuppositions and illocutionary operators behave differently in two other respects: presuppositions can be accommodated or corrected by an addressee, whereas illocutionary forces cannot be. Presuppositions can be corrected by e.g. metalinguistic negation (Horn 1989, Green 2000) when “[presuppositions] conflict with the demands of the context” (Potts 2005: 34).51 (12) is a (rather classical) example of presupposition correction.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 51

Potts (2005) uses the term ‘cancellation’ to cover both the presupposition correction and the presupposition accommodation, but I distinguish them.

! 100

(12) A: The King of France is bald. B: The King of France is not bald - There is no King of France!

By using the definite determiner the, speaker A introduces the presupposition that there is a unique King of France. This presupposition, however, is corrected by speaker B, because B knows that there is no King of France and, therefore, that the presupposition introduced by A’s utterance conflicts with B’s context. Similarly, presuppositions can be accommodated if the presupposition trigger is under the scope of a certain operator, such as a conditional. For example, the sentence John will quit smoking presupposes that John smokes. However, the conditional in (13), where the presupposition of the consequent appears in the antecedent, does not presuppose it.

(13)

If John smokes, he will stop smoking when he reads this. (Faller 2002: 118, ex. 86)

Unlike presuppositions, neither expressive meaning nor illocutionary meaning can be corrected. First, consider Potts’ (2007b) example of the expressive meaning introduced by bastard:

(14)

That bastard Kresge isn’t late for work. (#He’s a good guy.) (Potts 2007b: 170, ex. 8a)

In (14), the utterance that Kresge is a good guy conflicts with the expressive meaning of bastard, thus making the entire conversation infelicitous. This shows that the expressive meaning (or ‘conventional implicature,’ in Potts 2005) of bastard cannot be denied or corrected. ! 101

Similarly, illocutionary force cannot be accommodated by the contextual requirement. The following example is from Faller (2002: 118), where the Cuzco Quechua reportative evidential -si is analyzed as changing the illocutionary force of the prejacent.

(15)

Sichus ni-wa-rqa-n

Juan

hamu-na-n-ta

chay-qa,

If

Juan

come-NMLZ-3-ACC

this-TOP, Juan-TOP

say-1O-PAST1-3

Juan-qa

hamu-nqa-s come-3FUT-si p= ‘If I was told that Juan will come, then Juan will come.’ EV=

speaker was told that Juan will come. (Faller 2002: 118, ex. 87)

In (15), the antecedent of the conditional is the same as the evidential implication introduced by si- in the consequent. If the evidential implication introduced by -si- is a presupposition, it should be accommodated, as shown in (15). However, according to Faller (2002), (15) is felicitous, and the evidential implication in (15) is not accommodated, which indicates that -si- in Cuzco Quechua is an illocutionary operator.52 Examples such as (13) and (15) cannot be replicated in Korean, because the translation of (15) into Korean with reportative -ta- is unacceptable, possibly due to the ungrammaticality of embedding -ta- (I will discuss the embedding issue in the next section).

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 52

At the moment, it is unclear whether metalinguistic negation targeting the illocutionary operator is possible.

! 102

(16)

??John-i

Mary-lul

John-Nom Mary-Acc

ttayli-ess-ta-ko

han-ta-myen,

John-i

Mary-lul

beat-Past-Decl-Comp

say-Decl-Cond John-Nom Mary-Acc

ttayli-ess-ta-nta. beat-Past-ta-Decl Intended: ‘If it is said that John beat Mary, John beat Mary.’

However, we can still construct examples of presupposition correction involving the Korean reportative evidential -ta-. Consider the following conversation:

(17) A: John-i John-Nom

ecey

Mary-lul

manna-ess-ta-ni?

yesterday

Mary-Acc

meet-Past-ta-Q

‘Did John meet Mary?’ Implication: the addressee is expected to answer based on reportative evidence. B: Ung, Yes

sasil

ecey

nay-ka John-i

Mary-lul

actually

yesterday

I-Nom John-Nom

Mary-Acc

manna-nun

kes-ul

po-ass-e.

meet-Rel

thing-Acc

see-Past-Decl

‘Yes, actually I saw yesterday that John met Mary.’

Here, speaker A introduces the reportative evidential implication with -ta- that speaker B is expected to answer based on reportative evidence. However, B indicates that he/she has direct evidence rather than reportative evidence by saying that he/she saw that John met Mary, thereby cancelling the implication introduced by -ta- in A’s utterance. Examples such as this one indicate

! 103

that the implication introduced by -ta- is a presupposition, rather than a conventional implicature/expressive meaning or illocutionary meaning.53 Let us turn now to -te-. First of all, the implication introduced by -te- is not a conversational implicature. Consider examples such as (18), where the implication carried by the utterance containing -te- cannot be cancelled by the immediately following utterance (cf. the examples in fn. 5).

(18)

pi-ka

o-Ø-te-la.

Rain-Nom

fall-Pres-te-Decl

#Na-nun I-Top

pi-ka rain-Nom

o-nu-n

kes-ul

po-cianh-ass-ta.54

fall-Impf-Rel

thing-Acc

see-Neg-Past-Decl

‘[The speaker saw that] it was raining. #I did not see that it was raining.’ (Lee J. 2010: 13, ex. 43)

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 53

Matthewson et al. (2007) and Faller (2002, 2006) adopt the common conception of presuppositions and argue that with respect to ‘cancellability,’ two theories of evidentials—evidentials as epistemic modals triggering presuppositions and evidentials as illocutionary operators—converge. Thus, neither presuppositions nor illocutionary meanings can be cancelled. The following example from Matthewson et al. (2007) shows that, unlike conversational implicatures, presuppositions introduced by the St’át’imcet’s reportative evidential ku7 cannot be cancelled: (i)

#ts’um’-qs-án’-as ku7 kw s-Lémya7 Lick-nose-DIR-3ERG REPORT DET NOM-Lémya7 ats’x-en-lhkán wi7 zam’ see-DIR-1SG.SUBJ EMPH after.all ‘[reportedly] Lémya7 kissed roger; actually I saw it.’

kw DET

s-Roger; NOM-Roger

(Matthewson et al. 2007: 217, ex. 37) Compare (i) with the following example, where the conversational (scalar) implicature introduced by the numeral three is cancelled: (ii)

I have three kids. Actually, I have four, but the youngest is still a baby.

However, in my view, examples such as (i) show only that the implication introduced by ku7 is not a conversational implicature and, as shown above, there are several examples where presuppositions can be corrected or accommodated, which Matthewson et al. (2007) and Faller (2002, 2006) seem not to be considering. 54 The evidential implication is indicated in square brackets but does not mean that evidential markers are embedding predicates.

! 104

(19)

pi-ka

o-ass-te-la.

Rain-Nom

fall-Past-te-Decl

#Na-nun

pi-ka

o-ass-ta-ko

chwulonha-cianh-ass-ta.

I-Top

rain-Nom

fall-Past-Decl-Comp

infer-Neg-Past-Decl

‘[The speaker inferred that] it had rained. #I did not infer that it had rained.’ (Lee J. 2010: 13, ex. 44)

Like -ta-, -te- cannot appear in the consequent of a conditional (again, possibly due to the ungrammaticality of embedding -te-), meaning that conditionals such as (13) and (15) cannot be applied to Korean as a test for the presupposition accommodation, because the Korean translation of (15) with the direct evidential is unacceptable.

(20)

??Nay-ka John-i

Mary-lul

ttayli-nun-kes-ul

po-ass-ta-myen,

I-Nom John-Nom

Mary-Acc

beat-Past-Decl-Comp

see-Decl-Cond

John-i

Mary-lul

ttayli-te-la.

John-Nom

Mary-Acc

beat-Past-ta-Decl

Intended: ‘If I saw that John beat Mary, John beat Mary.’

However, the implication introduced by -te- can be challenged and corrected.

(21) A: Ecey Yesterday

pi-ka

o-Ø-te-la.

rain-Nom

fall-Pres-te-Decl

‘[The speaker saw that] it was raining yesterday.’

! 105

B: Melako? What do you mean

Camsiman.

Ne

ecey

kamok-ey

wait a minute

you

yesterday jail-Loc

pi-ka

o-nun kes-ul

po-ass-ullikaeps-e.

rain-Nom

fall-Rel thing-Acc

see-Past-Cannot-Decl

is-ess-nuntey, be-Past-so

‘What? Wait, you were in jail yesterday, so there’s no way you could have seen that it was raining.’ (Lee J. 2010: 13, ex. 41)

(22) A: Ecey Yesterday

pi-ka

o-Ø-te-la.

rain-Nom

fall-Pres-te-Decl

‘[The speaker saw that] it was raining yesterday.’ B: Melako?

Camsiman.

Ne ecey onul

kyeysok

What do you mean wait a minute you yesterday today continuously is-ess-nuntey,

mwusun

kunke-lo

kulehkey

be-Past-so

What

evidence-with so

kamok-ey jail-Loc

sayngkakha-ni? think-Q

‘What? Wait, you have been in jail continuously yesterday and today. What evidence makes you think so?’ (Lee J. 2010: 13, ex. 42)

This also shows that the implication introduced by -te- can be challenged and corrected if contextually required and, therefore, that it is presuppositional.

! 106

3.1.4. Interlude: the nature of the evidential presupposition

Up to now, I have argued that the evidential implications introduced by Korean -ta- and te- are presuppositional. One problem for our presuppositional analysis of Korean evidentials, however, is that they lack the distinguishing property of what is typically taken to be a presupposition. Although researchers disagree on the exact nature of presuppositions, the most widely accepted view follows Stalnaker (1974) in characterizing them as propositions that speakers implicitly present as taken for granted, that is, as beliefs that are shared among the participants in the conversation (modulo accommodation, see the discussion in Soames 1982). The evidential implication, which I described as a presupposition in this chapter, does not appear to have this property. Specifically, by employing an evidential marker, the speaker typically introduces new information, which does not need to follow from the conversational background. Consider the following conversation:

(23) A: Ecey Yesterday

John-i

mwusun

il-ul

ha-ess-ni?

John-Nom

what

thing-Acc

do-Past-Q

ecey

Mary-lul

manna-te-la.

yesterday

Mary-Acc

meet-te-Decl

‘What did John do yesterday?’ B: John-un John-Top

‘John met Mary yesterday.’ Implication: the speaker saw that John met Mary.

Here, A’s question gives no indication of his/her expectations concerning the type of evidence B’s answer may be based on. Therefore, A’s utterance is compatible with A’s complete neutrality ! 107

relative to B’s evidence, that is, with a conversational background that contains no information on the type of evidence available to B. This notwithstanding, B’s answer containing -te- is entirely felicitous and, therefore, has the effect of introducing the new information that he/she has direct evidence (or he/she directly saw) that John met Mary yesterday. At this point, we are facing a dilemma. On the one hand, indications introduced by Korean evidentials of the type of evidence available to the speaker behave exactly like presuppositions relative to projection and to the test of the truth or falsity of the prejacent. On the other hand, they appear to always lack the central characterizing property of presuppositions. This question can also be formulated as follows.55 In (23), suppose that B actually has no idea what John did yesterday and, therefore, that B has no evidence regarding whether John met Mary. However, B decides to lie and pretend that he/she saw that John met Mary. In this case, B may answer as in (23b). If his/her lie is successful, B is not saying something undefined or infelicitous, but suggests (deceitfully) that he/she has direct evidence that John met Mary yesterday.56 A plausible solution comes from Sauerland (2007) and Schlenker (2007), who note that presuppositions, if characterized as definedness conditions (Heim & Kratzer 1998), can also introduce new information by referring to context as defined in Kaplan (1989). In this chapter, I focus on Schlenker (2007). Reviewing Potts (2007b), which discusses the semantics of expressives such as damn and bastard, Schlenker hypothesizes that “expressives are lexical items !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 55

This example is from an anonymous reviewer’s comment on my abstract submitted for SALT 19. Aware of this problem, which is common to most presuppositional analyses of evidentials, Murray (2009a, b) suggests, in an analysis of evidential markers in Cheyenne, that one needs to distinguish two types of assertive meanings: non-at-issue assertion and at-issue assertion. An at-issue assertion is what we have been calling simply an ‘assertion,’ namely, a proposal to update common ground. Such assertions are negotiable (i.e., they may be challenged or denied). A non-at-issue assertion immediately updates the common ground without negotiation. According to Murray, Cheyenne evidentials contribute to non-at-issue assertions but do not affect at-issue assertions. This is intended to explain why Cheyenne evidential implications, like presuppositions, cannot be negotiated but, unlike presuppositions, introduce new information. Unfortunately, even if we were to posit this ‘dimension’ of meaning, Murray’s solution could not help us, because Korean evidential implications can be denied, corrected, or accommodated, as we have just seen in the previous subsection, and therefore, cannot be classified as non-at-issue assertions. In this sense, they are more similar to presuppositions, as formulated in this chapter. 56

! 108

that carry a presupposition of a particular sort” (2007: 237), which he names expressive presuppositions.57 According to Schlenker (2007), the main characteristics of expressive presuppositions are as follows:

(24)

An expressive presupposition is i) indexical: it is evaluated with respect to a context, and ii) attitudinal: it predicates something of the mental state of the agent of that context, and iii) shiftable: the context of evaluation need not be the context of the actual utterance. (Schlenker 2007: 237)

Schlenker (2007) further proposes that the meaning of expressives such as damn and bastard can be formulated in terms of expressive presuppositions, without assuming a separate semantic ‘dimension’ for the expressive meaning (contra Potts 2005, 2007b, 2007c). Specifically, he takes Stalnaker’s (2002) observation for granted, that is, “it is sometimes enough to present oneself as presupposing that p to guarantee that p is indeed common belief” (Schlenker 2007: 241), and further claims that the presuppositions that fit Stalnaker’s (2002) notion of expressive are those that predicate something about the speaker’s mental state. Since his expressive presuppositions are about the speaker’s mental state (see 22ii), they can add to the common ground simply by being presented. It is beyond the scope of this chapter to follow Schlenker’s (2007) argument (for a detailed formal explanation of how this can be done, the reader is referred to Stalnaker 2002 and Schlenker 2007). However, given that the properties of expressive presuppositions fit the facts of !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 57

As Schlenker (2007) points out, Soames (1989) also adopts the term expressive presupposition, defining it as follows: “A sentence S expressively presupposes a proposition P relative to a context C iff the truth of P is necessary for S to semantically express a proposition in C” (Soames 1989: 562). Schlenker (2007) treats his definition of the term as different from Soames’s (1989), saying that “[Soames’] use is now sufficiently rare that [my] title won’t create serious confusion” (2007: 237ff). Nevertheless, to avoid possible confusion, I use this term following Schlenker (2007) unless explicitly indicated otherwise.

! 109

Korean evidentials, I propose that Korean evidential markers introduce expressive presuppositions in the sense of Schlenker (2007). Importantly, however, I am not suggesting the semantics of Korean evidentials is identical to that of expressives.58 My point is weaker, namely, that evidentiality in Korean, like other implications in natural language, is presuppositional, and yet it can introduce new information to the common ground.

3.2. Korean evidentials and epistemic modals

In the previous section, I argued that the evidential implications introduced by Korean evidentials are presuppositions. In this section, I investigate whether, in addition to an evidential presupposition, Korean evidentials contribute an epistemic modal to the truth-conditions, as well. Apart from the conceptual proximity of the notions of evidentiality and epistemic modality, what motivates this inquiry is the fact, mentioned in Chapter 1, that other languages where evidentiality has turned out to be a presupposition have evidentials that are, in effect, epistemic modals carrying these presuppositions. Thus, it is reasonable to wonder whether Korean belongs to this set of languages. My conclusion is that there is NO empirical reason to assume that it does. I will argue for my claim with two tests: the assent/dissent test provided by Faller (2002), Papafragou (2006), and Matthewson et al. (2007), and the embeddability test mentioned in Matthewson et al. (2007) and applied to Korean by Lee J. (2010). Firstly, I will show that at least the test provided in Faller (2002) and Matthewson et al. (2007) is, unlike what they claim, not satisfactory to prove that there are components of epistemic modality in Korean evidentials. Secondly I will show that Korean evidentials become less acceptable when they are used in the embedded clause, unlike (epistemic) modals. My claim leads me to another conclusion that Korean evidentials belong to !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 58

As alluded in the introduction, in Chapter 4, we will also see the difference between expressives and evidentials, especially in questions.

! 110

the third type, which is not discussed in the introduction: they introduce evidential presuppositions, but do not contain any epistemic modal component.

3.2.1. Assent/dissent test59

As with the question addressed in the preceding sections, specific diagnostics have been suggested to address the question I am focusing on here: whether a given evidential marker contains components of epistemic modality or not. For example, Faller (2002) and Papafragou (2006) observe that, unlike an illocutionary operator, a truth-conditional modal operator can be the target of assent or dissent. This is because, in general, if a component of the meaning of a sentence is truth-conditional, it can be agreed upon or rejected, whereas if it is an illocutionary component, it cannot. Therefore, in order to detect an epistemic modal assertive component in evidentials, one needs to establish whether that component can pass the ‘assent/dissent’ test. Matthewson et al. (2007) further refine this test to distinguish the assertive modal component from both a presupposition and an illocutionary operator. Specifically, they claim that in St’át’imcets, a sentence containing an evidential can be followed by an expression of assent or dissent in light of the reliability of the source of evidence that the speaker based the utterance on. Given this, they conclude that this component of the meaning of evidentials (which can be characterized in terms of epistemic modality) is assertive in nature. (25) is an example of the St’át’imcets reportative evidential ku7 where the reliability of the source of information, which is the asserted part of the epistemic modal, is questioned.

(25)

(Context: Bill often lies. Yesterday, you heard him telling me that Buffy St. Marie is

coming to Mt. Currie to give a concert. That was the first you’d heard of it; you don’t !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 59

Thanks to Barry Schein and Stephen Finlay to point out problems of assent/dissent tests discussed in this subsection.

! 111

know whether it’s true or not, but think Bill’s probably lying. Then today, you hear me telling someone else…) cuz'

ku7

ts7as

going.to

REPORT come

k

Buffy St. Marie e-ts7á líl’wat-a

DET

Buffy St. Marie to-DEIC Mt. Currie-EXIS

“[reportedly] Buffy St. Marie is coming to Mt. Currie.” You say to me: aoz

kw

s-wenácw;

kakez7-úlh

k

Bill

NEG

DET

NOM-true

lie-always

DET

Bill

“That’s not true; Bill’s a liar.” (Matthewson et al. 2007: 224, ex. 54)

On the other hand, Matthewson et al. (2007) also claim that the component of an evidential that indicates the type of evidence source cannot be dissented in the same way is, so they analyze it as a presupposition. In (26), which also involves the St’át’imcets reportative evidential ku7, the presupposition about the information source ku7 introduces is dissented, resulting in infelicity.

(26)

(Context: your car was stolen) A: nilh FOC

ku7

s-Bill

REPORT NOM-Bill

ta

naq’w-ens-táli-ha

n-káoh-a

DET

steal-DIR-TOP-EXIS

1SG.POSS-car-EXIS

“[reportedly] It was Bill who stole my car.” B: #aoz NEG

kw

s-wenácw;

plan-lhkacw

lháp-en

kw

DET

NOM-true

already-2S.SUBJ forget-DIR

DET

s-7áts’x-en-acw

s-Bill

i

naq’w-ens-as

NOM-SEE-DIR-2s.conj

NOM-Bill

when.PAST

steal-DIR-3ERG

! 112

ti

káoh-sw-a

DET

car-2SG.POSS-EXIS

“That’s not true. You forgot you SAW Bill steal your car.” (Matthewson et al. 2007: 225, ex. 56)

At a first glance, it appears as if Korean -ta- and -te- behave similarly to the St’át’imcets evidentials. First, let us see -ta-. Consider (27), which is an application of Matthewson et al.’s (2007) assent/dissent test to the Korean non-final -ta- contains a component that can actually be assented/dissented.

(27)

(Scenario: Bill often lies. Yesterday Mary heard Bill telling John that Obama is coming to LA to give a speech. That was the first Mary had heard of it; she doesn’t know whether it is true, but thinks Bill’s probably lying. Today Mary heard John saying to Tom…) (adapted from Matthewson et al. 2007: 224) John:

Obama-ka

LA-ey o-n-ta-nta.

Obama-Nom

LA-Loc come-Pres-ta-Decl

‘(Reportedly) Obama is coming to LA.’ Mary: Ani, kukes-un sasil-i

ani-lke-ya.

Bill-un hangsang kecismal-ul ha-e.

No, that-Top fact-Nom Neg-may-Decl. Bill-Top always lie-Acc do-Decl ‘No, that may not be true - Bill always lies.’

In this scenario, Mary does not deny the prejacent of John’s assertion, but only expresses her skepticism relative to the reliability of John’s source of information. This example shows that this part of the semantic contribution of non-final -ta-, which can be described in terms of epistemic modality, can be challenged and, therefore that it contributes to the truth-conditional meaning. ! 113

Unlike the assertive component of -ta-, the presupposition of -ta- cannot be dissented (except through metalinguistic negation, which we saw in Section 3.1.3). Consider the following example, where the same scenario as that of (26) is applied to Korean -ta-:

(28)

(Scenario: your car was stolen) A: Bill-i Bill-Nom

nay cha-lul

hwumchi-ess-ta-nta.

my car-Acc

steal-Past-ta-Decl

‘(reportedly) Bill stole my car.’ B: #Kukes-un sasil-i

ani-ya.

Ne-nun

Bill-i

ney cha-lul

That-Top fact-Nom

Neg-Decl

you-Top

Bill-Nom

your car-Acc

hwumchi-nun-kes-ul

po-n-kes-ul

icepeli-ess-e.

steal-Rel-thing-Acc

see-Rel-thing-Acc

forget-Past-Decl

‘That’s not true. You forgot you saw Bill stole your car.’

As shown in (28), dissenting the information source results in an infelicitous conversation. This means that, unlike the reliability of the source, the type of information source is presuppositional, not truth-conditional. However, it is hasty to conclude that -ta- contains any component of epistemic modality because, whereas it is relatively clear that the presuppositional component of -ta- cannot be assented or dissented, as in (28), concerning the modal component, it is unclear what is actually assented or dissented. Specifically, the result of the test in (27) (and maybe in 26 as well) is compatible with dissenting the prejacent. For example, an utterance like (29), where the prejacent of the sentence with -ta- in (27) is directly dissented, can felicitously follow the conversation given in (27):

! 114

(29)

Mary: ... naykapokiey In my view

Obama-nun

LA-ey o-cianh-a.

Obama-Top

LA-Loc come-Neg-Decl

‘In my view, Obama does not come to LA’

In examples with -te-, we face the same problem. First, as in (28), it can be relatively easily observed that the evidential component of -te- cannot be dissented, as shown in (30) (again, except by metalinguistic negation, discussed in Section 3.1.3).

(30)

(Scenario: your car was stolen) A: Bill-i Bill-Nom

nay cha-lul

hwumchi-te-la.

my car-Acc

stole-te-Decl

‘(I saw that) Bill stole my car.’ B: #Kukes-un sasil-i That-Top fact-Nom

ani-ya.

Ne-nun

Bill-i

ney cha-lul

Neg-Decl

you-Top

Bill-Nom

your car-Acc

hwumchi-ess-ta-nun-kes-ul

Mary-eykey

tut-ess-ulppwun-i-ya.

steal-Past-Decl-Rel-thing-Acc

Mary-from

hear-Past-only-Cop-Decl

‘That’s not true. You only heard from Mary that Bill stole your car.’

We see in (30) that when the presupposition that -te- introduces about the information source is dissented, the sentence is infelicitous. This means that, as with -ta-, the evidential implication that -te- introduces is presuppositional. However, regarding the epistemic modal component of -te-, just like -ta-, the result of the test is not conclusive. At first sight it appears as if the reliability of the source encoded in -tecould be dissented, as in (31):

! 115

(31)

(Scenario: John’s friends believe that he cannot see well at night. However, John insists that he sees very well at night. Yesterday night, John saw someone stealing his car. Based on what he saw, he thought it was Bill who stole his car. Today, John met his friend Mary and said…)

John: (Nay-ka po-ass-nuntey) ecey I-Nom see-Past-so

pam-ey

yesterday night-Loc

Bill-i

nay cha-lul

Bill-Nom

my car-Acc

hwumchi-te-la. steal-Past-te-Decl ‘(I saw it - ) last night Bill stole my car. Mary: Hwaksilhay? Are-you-sure

Ne

pam-ey

nwun-i

cal

an

poi-ciah-a?

you

night-Loc

eye-Nom

well

Neg

see-Neg-Q

‘Are you sure? Don’t you not see very well in the night?’

Following Mattewson et al. (2007), one might be tempted to conclude that -te- contains components of epistemic modality. However, as we saw in cases of -ta-, Mary’s answer in (29) is still compatible with dissenting from the prejacent (that Bill stole John’s car last night). For example, Mary’s answer in (29) can be felicitously continued by (32) where the prejacent is challenged.

(32)

…Naykapokiey Bill-i In my view

Bill-Nom

ney cha-lul

hwumchi-chanh-ass-e.

your car-Acc

steal-Neg-Past-Decl

‘In my view Bill did not steal your car’

Concluding this subsection, we saw that the assent/dissent test given in Matthewson et al. (2007) is not enough to show that there are assertive components of epistemic modality in Korean ! 116

evidential markers. Therefore, I leave the possibility of the presence of epistemic modality in Korean evidentials open to further investigations, but for simplicity, I do not include the epistemic modal component in the lexical entry for Korean evidentials below.60

3.2.2. Embeddability

Faller (2002) as well as Matthewson et al. (2007) mentions embeddability as distinguishing epistemic modals (and evidentials whose semantics is that of epistemic modals) from illocutionary operators. As shown in (33), in general, epistemic modals can appear in an embedded proposition (even though the type of matrix predicates is limited: see Papafragou 2006):

(33)

John believes that Mary must be a genius.

However, -ta- generally resists appearing in the embedded proposition. (34), where -ta- appears in the embedded proposition, is simply unacceptable for most Korean speakers, no matter what the matrix predicate might be, which is unexpected if -ta- is an epistemic modal:

(34)

*John-un John-Top

Mary-ka

Bill-ul

manna-ess-ta-nta-ko

Mary-Nom

Bill-Acc

meet-Past-ta-Decl-Comp

tut-ess-ta / malha-ess-ta / mit-ess-ta... hear-Past-Decl / say-Past-Decl / believe-Past-Decl

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 60

In the next chapter we will see how the lexical entries for Korean evidentials look like when the component of epistemic modality is included.

! 117

Regardng the direct evidential -te-, judgments diverge. Lee J. (2010) reports that examples like (35) and (36), where -te- is in the embedded proposition, are acceptable. Specifically, she claims that, when -te- appears in the embedded proposition whose matrix predicate is malha- ‘say’, there is an evidential implication such as the subject of the matrix clause, not the utterer, has direct evidence about the truth or falsity of the embedded proposition.

(35)

(Scenario: John mistakenly left his cell phone in his office yesterday. When he got back to the office late at night, the door was not locked) John-un ecey

cenyek-ey samwusil mwun-i

John-Top yesterday evening-at office door-Nom onul achim-ey

malha-ess-ta.

today morning-at

Say-Past-Decl

yellyeiss-+-te-la-ko be.open-Pres-te-Decl-Comp

‘Today morning John said that [John saw that] the office door was open yesterday evening.’ (Lee J. 2010: 14, ex. 47)

(36)

(Scenario: Yesterday John found out that Mary’s blanket smelled good, and all stains were removed) John-un ecey

Mary-ka

tamyo-lul

ppal-ass-te-la-ko

John-Top yesterday

Mary-Nom

blanket-Acc

wash-Past-te-Decl-Compl

onul achim-ey malha-ess-ta. this morning-at say-Past-Decl ‘Today morning John said that [John inferred that] Mary washed the blanket yesterday.’

! 118

However, for me (35) and (36) are just unacceptable. When they are acceptable (and when they have the implication which Lee J. claims there is), it is only when the embedded proposition is a direct quotation of what John said yesterday, which is clearly different from the usual cases of embedding. This possibility of confusion between direct quotation and embedding makes Lee J.’s (2010) claim that -te- is embeddable less compelling. To sum up, just like the assent/dissent test, the embeddability test does not show that Korean evidentials are epistemic modals. Rather, evidence seems to point the opposite direction: that is, Korean evidentials are, even though they are presupposition triggers (like evidentials in some languages which are also epistemic modals), they behave in a different way from that of epistemic modals, and therefore not epistemic modals.

3.3. Summary and conclusion

In this chapter, I reviewed the semantic characteristics of Korean evidentiality. I argued that Korean evidential markers introduce evidentiality as presuppositions, but there is no compelling evidence supporting the claim that Korean evidentials are epistemic modals (contra Lee J.’s claim for Korean direct evidential -te- in Lee J. 2010). Typologically, as pointed out above, this claim also implies that Korean is a language whose evidentials do not belong to any type introduced in Chapter 1, but belong to the third type: they are neither epistemic modals nor illocutionary operators, but are mere presupposition triggers which do not contain any assertive component. In the following appendix, I will illustrate one difference of -ta- and -te regarding the variability of the degree of the speaker’s certainty on the prejacent, and review a recent proposal made in Matthewson et al. (2007) about the similar variability in St’át’imcets evidentials. ! 119

3.4. Further discussion: speaker’s certainty and modal quantificational force

In this section I will discuss the (apparent) variability of the degree of certainty in Korean reportative evidential -ta- and St’át’imcets evidentials. To do that, let us begin by illustrating how the standard Kratzerian semantics of (epistemic) modals can be applied to the evidentials whose semantics is that of epistemic modals with evidential presuppositions. In Kratzer’s semantics of modality (e.g., Kratzer 1991), modals are treated as quantifiers over possible worlds, either existentially or universally. Existential quantification and universal quantification over possible worlds in the modal base correspond to the possibility interpretation and the necessity interpretation of modality, respectively, and different types of quantification correspond to different degrees of speaker certainty about the truth of the embedded claim. For an illustration of how this can be applied to the semantics of evidentials, let us return to the perfective evidentials (PEs) in Izvorski (1997). Izvorski’s definition of PE is as follows:

(37)

The Interpretation of PE(p): a. Assertion: !p in view of the speaker’s knowledge state b. Presupposition: Speaker has indirect evidence for p (Izvorski 1997: 226, ex. 8)

Under this definition, a PE has the same semantics as the English epistemic modal must, except that it introduces an indirect evidential presupposition. Only the possible worlds where the evidential presupposition holds remain in the modal base available to the speaker. In addition to this, Kratzer’s (1977, 1991) semantics of modality assumes that the ordering source, which is also ! 120

determined depending on the speaker’s context, further restricts the possible worlds in the modal base. Let me briefly illustrate how Izvorski’s proposal explains examples such as (38).

(38)

Ivan

izpil

vsi#koto

vino

v#era.

Ivan

drunk-PE

all-the

wine

yesterday.

‘Ivan apparently drank all the wine yesterday.’ (Izvorski 1997: 228, ex. 12)

Suppose that the speaker of (38) heard from Mary that Ivan drank all the wine yesterday.61 In this case, the speaker has reportative evidence that Ivan drank all the wine yesterday. Because of the presupposition of the PE, the modal base, which the PE universally quantifies over, includes only possible worlds where the proposition that Mary said that Ivan drank all the wine yesterday is true. To formalize this intuition, let B(w) be the modal base for the PE in the world of evaluation w. Then, roughly speaking, B(w), the set of possible worlds, can be formalized as follows:62

(39)

B(w) = {w: Mary said that Ivan drank all the wine yesterday in w}

That is, B(w) includes all possible worlds where the proposition that Mary said that Ivan drank all the wine yesterday is true. The modal base in (39) is not sufficient to derive the meaning we have for (33), however, because it may include possible worlds that are not contextually relevant. For instance, we may imagine a possible world w’ where it is the case that Mary said that Ivan drank all the wine !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 61

According to Izvorski (1997), PEs may indicate that the speaker has either reportative or inferential evidence. For convenience, I only discuss the reportative interpretation of PEs, but my analysis can easily be extended to the inferential interpretation of PEs. 62 This formalization is obviously very sloppy. For a strict formalization of epistemic modals that this thesis adopts, see Kratzer (1977, 1991), and for a detailed explanation of Kratzer’s semantics of modals, see Portner (2009).

! 121

yesterday, but where it is also the case that Mary always lies and, therefore, is unreliable. The set of possible worlds, which the PE (and modals in general, actually, in Kratzer’s semantics of modals) quantifies over, should rule out this kind of possible world. This means that the modal base should be further restricted, and this can be done by what Kratzer (1977, 1991) calls the ordering source, along which possible worlds in the modal base are ordered with respect to how similar each is to the ‘ideal’ world the speaker has in mind. In Kratzer (1977, 1991), the ordering source is defined as a set of propositions, all of which are true in the speaker’s ‘ideal’ world. Let us assume that, in (38), the ordering source O in the world of evaluation w is a singleton set, the element of which is the proposition q that Mary is a reliable source of information in w. Then, the ordering source can be formalized in terms of a set as in (40).

(40)

O(w) = {p: p= %w. Mary is a reliable source of information in w}

Given (40), then for the set of all possible worlds W and the conversational background B(W), the partial ordering with respect to O(w), &O(w), can be formalized as follows (Kratzer 1991: 644, definition 5):

(41)

For all w', w'' ' W, for any O(w) % B(W): w'&O(w)w'' iff {p: p'O(w) and w'''p} % {p: p'O(w) and w’ ' p}

According to this definition, “a world w' is at least as close to the ideal represented by O(w) as a world w'' iff all propositions of O(w) which are true in w'' are true in w' as well” (Kratzer 1991: 644).

! 122

With respect to these two contextually determined parameters - the modal base B(w) and the ordering source O(w) - the meaning of a PE can be defined as follows (Izvorski 1997: 230, ex. 19):

(42)

[[PE(p)]]B,O,w = {w'W: &u'W[(u'B(w) " ~!v'W (v'B(w) v
According to this lexical entry, PE(p) is true with respect to the modal base B and the ordering source O if and only if, among all worlds in B where the speaker’s indirect evidence holds, in the worlds closest to the ideal world provided by O, the prejacent is true. In terms of (38), (38) is true if and only if, for all worlds where it is true that Mary said that John drank all the wine yesterday, it is true that John drank all the wine yesterday. Izvorski’s (1997) analysis of PE cannot simply be extended to evidentials in St’át’imcets, which can also be epistemic modals with evidential presuppositions. This is because, as observed by Matthewson et al. (2007), a simple-minded analysis of evidentials in terms of universal epistemic modality fails to capture the variation that evidentials can sometimes allow relative to the speaker’s level of certainty. Specifically, Matthewson et al. (2007) observes that the St’át’imcets reportative evidential ku7 is felicitous when the source of evidence is very reliable, such that the speaker has a very high degree of certainty with respect to the prejacent, as in (43), but that it can also be used felicitously when the speaker’s certainty is less strong, as in (44).

(43)

(Scenario: the speaker would like to say that his/her father drives a cab, but the speaker has not seen it directly, but only heard it from someone else who he/she thinks is very reliable…) wa7 ku7

aylh múta7 tq-álk’-en-as

IMPF REPORT

then and

ta

taxicab-a

knáti7 táown-a

tough-string-DIR-3ERG DET taxicab-EXIS DEIC

town-EXIS ! 123

“[reportedly] he [my father] also drove a taxicab around town.” (Matthewson et al. 2007: 238, ex. 86)

(44)

(Scenario: There is a rumor going around that Roger was elected chief. Sometimes, that kind of rumor is right, but sometimes it’s wrong. You have no idea whether it’s right or wrong. You tell me…) %

aw-an-ém

ku7

Choose-DIR-PASS

kw

REPORT DET

s-Roger

ku

cuz’

kúkwpi7

NOM-Roger

DET

going.to chief

“[reportedly] Roger was elected to be chief.” (Matthewson et al. 2007: 239, ex. 87)

According to Matthewson et al. (2007), other evidential markers in St’át’imcets show similar quantificational variability. Turning now to Korean, we find that, whereas the direct evidential -te- generates exactly the abovementioned prediction, a simple-minded universal analysis of the modal component of the reportative evidential -ta- presents the same problem as the evidentials in St’át’imcets. In fact, while -te- is compatible only with a high degree of speaker certainty regarding the truth of the prejacent, as shown in (45) and (46), -ta- is legitimate whether the speaker is very certain or completely skeptical about the truth of the prejacent, as shown in (47) and (48).

- Examples with -te-

(45)

John-i

ecey

John-Nom yesterday

Mary-lul

manna-te-la.

Nay-ka

po-ass-nuntey,

Mary-Acc

meet-te-Decl

I-Nom

see-Past-Conj

! 124

thulimeps-e. it-must-so-Decl ‘[I saw that] John met Mary yesterday. I saw it, and it must so.’

(46)

John-i

ecey

John-Nom yesterday

Mary-lul

manna-te-la.

#na-nun

ku sasil-i

Mary-Acc

meet-te-Decl

I-Top

that fact-Nom

uysimsulep-e. doubtful-Decl ‘[I saw that] John met Mary yesterday. #I am doubtful of that fact.’

- Examples with -ta-

(47)

Ecey

pi-ka

o-ass-ta-nta.

Kuliko na-nun kukes-ul

Yesterday

rain-Nom

come-Past-ta-Decl

and

hwaksilhi

mit-e.

certainly

believe-Decl

I-Top

it-Acc

‘(Reportedly) it must have rained yesterday, and I certainly believe it.’

(48)

Ecey

pi-ka

Yesterday rain-Nom

o-ass-ta-nta.

Haciman na-nun kukes-i uysimsulep-e.

come-Past-ta-Decl

but

I-Top

it-Nom doubtful-Decl

‘(Reportedly) it may have rained yesterday, but I doubt it.’

On the one hand, the sentence with -te- can be followed felicitously by an utterance expressing that speaker is fairly sure about what he/she directly perceived, but not by an utterance expressing the speaker’s doubt. These two examples show that when using -te-, the speaker’s degree of ! 125

certainty is almost as high as when using a necessity modal such as must. On the other hand, both (47) and (48), which contain -ta-, are felicitous, which suggests that the speaker’s degree of certainty regarding the prejacent is variable when -ta- is used; the speaker may be strongly certain that the prejacent is true or may seriously doubt it. This difference might appear natural, since direct perception, e.g., vision, is usually regarded as one of the most reliable sources of information, but a straightforward universal modal semantics for both evidentials will not account for it. If -ta- is indeed an epistemic modal, then it appears to be one that exhibits a context dependent quantificational variability. Matthewson et al. (2007) suggest that the apparent quantificational variability of the St’át’imcets evidentials can be accounted for if one adopts Klinedinst’s (2005) theory of possibility modals, according to which possibility modals are similar to plural indefinite DPs. Thus, just like plural indefinites, which existentially quantify over plural entities, possibility modals existentially quantify over pluralities of possible worlds. Specifically, a set of possible worlds, the membership of which might be plural, is existentially quantified over, and then the individual worlds in that set are universally quantified over. Klinedinst’s (2005) lexical entry for the possibility modal is given below.

(49)

MODAL(p)

is true with respect to a modal base B and a possible world w iff:

!W[W%B(w) " W,Ø " &w’[w’'W*p(w’)]] (Matthewson et al. 2007: 240, ex. 88)

Informally, a modal is true of p if “there is a set of worlds W that are accessible from w [the world of evaluation], such that p is true in every world in W” (Matthewson et al. 2007: 241). Matthewson et al. (2007) and Rullmann et al. (2008) further revise Klinedinst’s (2005) proposal, arguing that the St’át’imcets modals are analogous to specific plural indefinites (“there ! 126

is a specific set of worlds W…”). In order to analyze epistemic modals as quantifiers over specific sets of possible worlds, they suggest that their lexical semantics includes references to a choice function that returns as the domain of universal quantification a subset of the set of worlds accessible from the world of evaluation. The lexical entry they suggest for the St’át’imcets modals is given below.

(50)

[[modal]]c,w is only defined if c provides a modal base B. If defined, [[modal]]c,w = %f. %p.&w’[w’'f(B(w))*p(w’)] (Matthewson et al. 2007: 241, ex. 89)

According to this lexical entry, modals take as their argument a choice function and universally quantify over the subset of the modal base selected by that choice function. The choice function is assumed to be provided contextually and, therefore, may vary across contexts. As a result of this cross-contextual variation, larger or smaller subsets of the modal base are quantified over. Clearly, the wider the set of worlds constituting the domain of the universal quantifier, the higher the speaker’s degree of certainty. If the speaker is absolutely certain of p, the choice function f will be the identity function, meaning that every world in the modal base is universally quantified and in all the worlds accessible from the world of evaluation, p is true. If the speaker is less certain about p, the choice function f selects fewer possible worlds from the modal base (and, therefore, f(B(w)) becomes smaller). Therefore, Matthewson et al. (2007) and Rullmann et al. (2008) encode the variability in the speaker’s degree of certainty regarding the asserted proposition in the contextual variation of the choice function. Matthewson et al. (2007) describe the lexical entries for the St’át’imcets evidential markers in terms of (44). They only differ depending on the type of evidence that holds in the

! 127

modal base accessible from the world of evaluation. For example, the lexical entries for the St’át’imcets inferential evidential k’a and reportative evidential ku7 are given in (45).

(51) a. [[k’a]]c,w is only defined if c provides a modal base B such that for all worlds w’, w’'B(w) iff the inferential evidence in w holds in w’. If defined, [[k’a]]c,w = %f. %p.&w’[w’'f(B(w))*p(w’)]] b. [[ku7]]c,w is only defined if c provides a modal base B such that for all worlds w’, w’'B(w) iff the reportative evidence in w holds in w’. If defined, [[ku7]]c,w = %f. %p.&w’[w’'f(B(w))*p(w’)]] (Matthewson et al. 2007: 243, ex. 91)

However, the sort of data Matthewson et al. (2007) presented does not require any quantificational variability of epistemic modals as Matthewson et al. (2007) proposes.63 Consider following examples (Barry Schein p.c.):

(52) a. My daddy drives a cab. It was (even) reported that he drove one around town. b. It was reported that Roger was elected chief, but can you always believe what you hear?

In (52a) the speaker asserts that his/her daddy drives a cab (and obviously he/she strongly believes that it is true), and a report from someone else does not affect this certainty, whereas in (52b) the speaker does not believe what he/she has heard. This does not change if we change the embedding clause it was reported to the corresponding sentential adverb reportedly. Importantly, both the expression it was reported and the sentential adverbial reportedly do not encode the degree of the speaker’s certainty, but simply introduces the implication (no matter what it is truth!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 63

Thanks to Barry Schein to point out these problems.

! 128

conditional or presuppositional) that there is a report from someone else about the speaker’s assertion. Even though these two examples do not use any evidential marker (in the narrow sense), they suggest that the apparent variability of the degree of certainty found in St’át’imcets evidentials and Korean -ta-.64 To sum up, I do not adopt Matthewson et al.’s (2007) proposal to analyze the variability of certainty of -ta-. In my view, this variability is more likely to be accounted for in terms of the pragmatics of judging the liability of reports: Korean -ta- simply indicates that the speaker has reportative evidence relative to the prejacent, and does not contain any component related to the speaker’s degree of certainty.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 64

Portner (2009) pointed out that Matthewson et al. (2007) and Rullmann et al. (2008) do not make overt use of any ordering source, and if we adopt the notion of the ordering source, there seems to be no difference in empirical predictions between the choice function and the ordering source. In Kratzer’s (1977, 1991) semantics of modality, the ordering source determines which possible worlds in the modal base fit the conversational background better than others. In this sense, the ordering source also ‘selects’ a set of possible worlds from the modal base, as do the choice functions assumed in Matthewson et al. (2007) and Rullmann et al. (2008). This means that, since both an ordering source and a choice function are determined based on context, there is no empirical difference between the (rather standard) Kratzerian semantics of modality and Matthewson et al.’s (2007) and Rullmann et al.’s (2008) semantics of modality. If Portner (2009) is correct, the facts regarding St’át’imcets evidentials can be accounted for by simply arguing that the “St’át’imcets evidentials are necessity modals, and the range of ordering sources compatible with them is different from those familiar from analyses of English, German, and other languages” (Portner 2009: 142). The problem, however, is that if we attribute the quantificational variability of the St’át’imcets evidentials to the ordering source, it may make incorrect empirical predictions about epistemic modals in languages where epistemic modals do not show quantificational variability. Consider English must, for example. When used epistemically, must always expresses the speaker’s strong certainty on the prejacent. Since, under a Kratzerian semantics of modals, must is also analyzed in terms of the modal base and the ordering source, if the variability in the speaker’s certainty is due to the ordering source, we predict that English must should also show variability according to the speaker’s degree of certainty, contrary to fact. Therefore I do not adopt Portner’s (2009) account here.

! 129

Chapter 4 Questions as sets of characters

This chapter concerns the interpretation of interrogative sentences containing reportative evidentials in Korean. In particular, this chapter focuses on what has come to be known as the ‘origo shift of evidentials’ in questions (Garrett 2001), which is referred to here as the ‘evidential perspective shift.’ The evidential perspective shift consists of an evidential being interpreted as regarding the answerer’s information source, rather than the questioner’s. Here, I introduce a novel semantic approach to this phenomenon that derives it as a consequence of the interaction of the meaning of questions and the meaning of evidentials. I also illustrate the conceptual and empirical advantages of this view over previous accounts developed in terms of pragmatics. Furthermore, I discuss the implications of the analysis proposed for the semantics of interrogatives and of evidentiality in general. As in previous chapters, I mainly focus on the Korean reportative evidential-ta-, briefly discussing the direct evidential -te- at the end of the chapter.

4.1. Introducing the puzzle

4.1.1 Non-final -ta- as reportative evidential

Let us recall the Korean morphological paradigm and Korean reportative evidentiality. As we saw in Chapter 2, in Korean, different sentential endings are employed depending on the register and the relative social status of the speaker and addressee in a given conversational ! 130

context. Specifically, when the speaker and addressee are of the same social status, the register is that of plain informal speech, and what is uttered is a declarative sentence, the speakers can choose the sentential ending -ta or -tanta, as shown in (1).

(1) a. John-i John-Nom

na-lul

po-ass-ta.

I-Acc

see-Past-Decl

na-lul

po-ass-ta-nta.

I-Acc

see-Past-ta-Decl

‘John saw me’ b. John-i John-Nom ‘John saw me’ Presupposition: The speaker has heard from someone else that John saw the speaker.

(1) shows, as several authors have observed (e.g., Kim N.-K. 1990), that the ending -tanta introduces a reportative evidential presupposition, but the declarative ending -ta1 does not. I have argued that these two endings are related in the following way: -tanta is of a combination of the declarative ending -nta, which is an allomorph of the declarative ending -ta in (1a) (hereafter -ta1), and a non-final -ta- (hereafter -ta2-), which carries evidentiality. As the reader certainly recalls from Ch. 2, in questions, too, two interrogative endings can occur in the Korean plain speech style when the speaker and addressee are of the same social status: -ni and -tani. Like its declarative counterpart, the ending -tani, which contains a non-final ta2-, introduces a reportative evidential implication, whereas the simple ending -ni does not. The implication in this case is that the speaker expects the addressee to answer based on reportative evidence.

! 131

(2) a. John-i John-Nom

na-lul

po-ass-ni?

I-Acc

see-Past-Q

‘Did John see me?’ b. John-i John-Nom

na-lul

po-ass-ta-ni?

I-Acc

see-Past-Decl-Q

‘Did John see me?’ Presupposition: The speaker expects the addressee to answer on the basis of reportative evidence.

Turning to other speech styles, we observe the same contrast as above, a parallelism previously ignored. For example, in the formal register, when the speaker is of lower social status than the addressee, the simple declarative ending is -(su)pnita and the simple question ending is (su)pnikka. In the informal register, when the speaker and addressee are in a close relationship (e.g., they are close friends or relatives, etc.), the simple declarative ending is -e/a and the simple question ending is -e/a (with a rising tone). The paradigm in (3) shows that for each of these morphologically simple endings, there is a complex counterpart that introduces a reportative evidential implication, where a non-final -ta2- precedes the corresponding simple ending.

(3)

Register

Simple (without evidentiality)

Complex (with evidentiality)

Declarative

Question

Declarative

Question

Formal

-pnita

-pnikka

-ta-pnita

-ta-pnikka

Plain

-(n)ta

-ni

-ta-nta

-ta-ni

Intimate

-e/a

-e/a

-ta-e/a (=> tay) -ta-e/a (=> tay)

! 132

This systematic correspondence between the presence of non-final -ta- and the implication that the speaker has reportative evidence about what he/she is asserting cannot be a mere coincidence. Based on this observation, in Chapter 2 I have concluded that this morpheme is different from declarative -ta1 and is responsible for the evidential meaning. Therefore, in what follows, I refer to this morpheme as -ta2 -, so as to distinguish it from the declarative simple sentential ending, which I refer to as -ta1.65 Lastly, recall that I also concluded, in Chapter 3, that the implication introduced by non-final -ta2- is a presupposition.

4.1.2 Context dependency of non-final -ta- and indexicality

Apart from the abovementioned observation on non-final -ta-, we also observed in Chapter 1 that, quite straightforwardly, the semantics of evidentiality must contain a component of indexicality. Since evidential -ta- refers to the source of evidence of whoever utters the sentence hosting it, its semantic behavior is clearly reminiscent of that of indexicals such as I, you, here, now, etc. Consider (4), repeated from (1b).

(4)

John-i

na-lul

po-ass-ta-nta.

John-Nom

I-Acc

see-Past-ta-Decl

‘John saw me’ Presupposition: The speaker has heard from someone else that John saw the speaker.

In (4), the evidential implication introduced by -ta- is speaker-dependent. Suppose that (4) were uttered by Mary. The evidential implication in (4) would then become ‘Mary has heard from !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 65

Hereafter, for convenience, I refer to Korean non-final -ta2- simply as -ta- unless there is confusion, since -ta1 is not particularly related to our main discussion.

! 133

someone else that John saw her (Mary).’ Similarly, if (4) were uttered by Bill, the evidential implication in (4) would become ‘Bill has heard from someone else that John saw him (Bill).’ In this sense, the reportative evidential implication introduced by -ta2- varies over utterance contexts in the same way as the reference of indexicals.

4.1.3. The “evidential perspective shift” puzzle

Granted that Korean non-final -ta- is a reportative evidential morpheme and, as such, should be analyzed as an indexical, its semantics in interrogatives presents an interesting puzzle, which I label the evidential perspective shift in questions (the origo shift in Garrett 2001). A comparison between (1b) and (2b), repeated below as (5a) and (5b), respectively, illustrates the puzzle.

(5) a. John-i John-Nom

na-lul

po-ass-ta-nta.

I-Acc

see-Past-ta-Decl

‘John saw me.’ Presupposition: The speaker has heard from someone else that John saw the speaker b. John-i John-Nom

na-lul

po-ass-ta-ni?

I-Acc

see-Past-ta-Q

‘Did John see me?’ Presupposition: The speaker expects the addressee to answer based on his/her reportative evidence whether John saw the speaker or not.

Evidentials and indexicals pattern very differently in questions than in declaratives (see 5a). Specifically, in the context of questions, the interpretation of evidentials shifts from the ! 134

perspective of the speaker (i.e., the questioner) to that of the addressee (i.e., the answerer), but the interpretation of other indexicals does not. In (5a), the evidential implication is that the speaker has reportative evidence that John saw the speaker, but in (5b), it is that the addressee is expected to have reportative evidence about whether John saw the questioner. If the evidential perspective in questions were fixed like that of indexicals, the evidential in (5b) would indicate that the speaker (i.e., the questioner) heard from someone else whether John saw the questioner, contrary to what we observe. Clearly, the interpretation of indexicals in questions does not shift in the same way. If Mary asked Tom the question in (5b), and the referent of “me” were fixed to the answerer’s (i.e., Tom’s) perspective, the question would mean ‘Did John see the answerer?’, that is, ‘Did John see Tom?’, contrary to fact. The main goal of this chapter is to explain the difference in (5) in a way that maintains the status of evidentials as indexicals, as they appear to be in declarative environments.

4.1.4. Outline of the chapter

In this chapter, I propose an explicit semantic account for the evidential perspective shift in questions in Korean. The account is based on the assumption that the Korean reportative evidential -ta- denotes a function from propositions (type ) to characters (type ) that introduces the evidential expressive presupposition that the speaker of the context variable bound in the character has access to indirect evidence regarding the argument proposition.66 A desirable consequence of this analysis is that, unlike other indexicals, an evidential can, in principle, be anchored to a context different from that of the utterance. This allows me to derive the evidential

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 66

I already outlined the notion of expressive presuppositions I am adopting here in Chapter 3, but for the detail, see Schlenker (2007) and Stalnaker (2002) which Schlenker adopts. See also Soames (1989) for a different definition of expressive presuppositions.

! 135

perspective shift in questions from the semantics of evidentials and its interaction with a standard semantics of questions (that in Hamblin 1973) without further semantic stipulation. The proposal outlined above has interesting implications relative to the type of semantic objects that questions can denote. Specifically, I show that a standard semantic account of questions, such as Hamblin’s (1973), predicts that questions containing -ta- denote sets of characters, whereas those without -ta- denote sets of propositions. The paper discusses how this result essentially captures the intuition already found in Garrett (2001), namely, that for the evidential shift to be possible, the denotation of a question must encode the possibility of anchoring to the addressee’s parameter. Unlike Garret’s proposal, however, this result is obtained here without ad hoc stipulations concerning the semantic composition of interrogatives. This chapter is organized as follows. Section 4.2 introduces the background assumptions I adopt regarding the semantics of indexicality and provides a brief overview of Hamblin’s semantics of questions. In section 4.3, I illustrate my proposal. Section 4.4 discusses and discards the possibility of an alternative account based on pragmatic factors alone. In section 4.5, I compare existing approaches to the perspective shift with the present proposal. Section 4.6 compares the evidential implication introduced by -ta2- with the implications introduced by other perspective-related items, such as expressives (cf. Potts 2005, 2007 and Harris & Potts 2010) and utterance-modifying adverbs (cf. Garrett 2001, Potts 2005, and Ernst 2009). Section 4.7 discusses the direct evidential -te-, and Section 4.8 concludes the chapter. Section 4.9 is a further discussion, where I will compare Hamblin’s (1973) semantics of questions with Karttunen’s (1977), especially with respect to the semantics of evidentials discussed in this chapter.

! 136

4.2. Kaplan’s theory of Indexicals and Hamblin’s semantics of interrogatives

This section summarizes the semantic theories of indexicality and interrogatives that I adopt in this chapter. For the semantics of indexicals, I adopt Kaplan’s (1989) two-step semantics, and for the semantics of questions, I follow Hamblin (1973). The reader who is already familiar with these two theories can skip this section and turn to section 4.3.

4.2.1. Kaplan (1989): pronouns as indexicals

Kaplan (1989), observing that natural language expressions such as here and now (a.k.a. indexicals) are assigned world-independent but context-dependent denotations, proposes a theory in which the referent of an indexical, once established in a given utterance context, behaves like a rigid designator across evaluation worlds (Kripke 1980), unlike, say, a definite description, whose referent ultimately depends on the extension of its restrictor in any given evaluation world. Kaplan’s (1989) proposal is intended to account for differences such as the contrast below (see also Soames 1989 for a brief introduction to this issue).

(6)

a. He must be rich. b. The President of GM must be rich.

Suppose that Mary utters (6a), intending the reference of the pronoun to be John. In such a case, (6a) is true if and only if, in every one of Mary’s epistemically accessible worlds (i.e., the epistemic alternatives she entertains), John is rich, that is, the referent of he is invariant across evaluation worlds and depends only on the speaker’s context. ! 137

Now consider (6b), where he is replaced by the definite description the President of GM. When (6b) has a so-called de dicto reading, that is, when the definite description takes narrow scope with respect to must,67 (6b) is true if and only if, in every one of the speaker’s epistemic alternatives, the president of GM, whoever that might be in that world, is also rich in that world, whereas the richness of any specific individual, such as John, in these worlds is not necessarily relevant to the truth of the sentence, unlike in (6a). Suppose that Mary utters (6b), and for simplicity, assume that Mary’s epistemic alternatives are limited to three worlds: w1, w2, and w3. Suppose also that in w1 John is the president of GM, and in w2 Jane is, and in w3 Bill is. In that case, (6b) is true if and only if John is rich in w1, Jane is rich in w2, and Bill is rich in w3, regardless of whether the others are rich in the other worlds. To capture this difference between indexicals and other descriptions, Kaplan (1989) proposes the following two-step semantics: In the first step, the reference of demonstratives/pronouns is determined depending on the context, and in the second step, the denotation of world dependent expressions, such as predicates, is established depending on the world of evaluation. More formally, in this framework, a sentence is generated by the grammar, and its meaning is called a character, that is, a function from contexts to intensions. In the first step, the character applies to a given context and returns the intension of the sentence relative to that context. In the second step, the intension applies to a world of evaluation and returns the extension of that sentence relative to that context and that world.

(7)

(revised from von Fintel 2005) a. The character of (:

[[(]]) = %c.%w.[[(]]c,w

b. The intension of ( at context c: [[(]]c¢ = %w.[[(]]c,w

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 67

I do not discuss the de re interpretation of (5b), which is not relevant to the discussion in this subsection.

! 138

Figure 4.1. General schema of Kaplan’s (1989) two-step semantics

Grammar ~~~> Sentence --------> Character --------> Intension --------> Extension * * Context World-of-evaluation (revised from von Fintel 2005)

Following common practice,68 in this chapter, the utterance context c is regarded as a triple, including the world of the utterance wc, time of the utterance tc, and speaker of the utterance sc. In addition to this parameter, the denotation of each expression is further relativized to the world of evaluation, which can be bound under intensional operators or remain free and ultimately be identified with wc.

(8)

Utterance context c = 69 Evaluation parameters: the world of evaluation w

To see how this applies to derive the difference observed in (6), consider (9).

(9) a. I am rich. b. [[I]]c* = [[I]] = s* c. If s* = Mary then [[I]]c* = [[I]] = Mary d. If s* = Tom then [[I]]c* = [[I]] = Tom

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 68

See Kaplan (1989: esp. Section XVIII) for a detailed formal system of indexicals and demonstratives. See also von Fintel (2005) for a brief summary of the Kaplan’s formalization, and the references therein for further discussion. 69 We only indicate the three parameters of context that are minimally required. We may also need the parameter of place of utterance or other parameters that are required to interpret other types of indexicals. For our purposes, however, these parameters of context are sufficient. We also ignore time of evaluation for simplicity.

! 139

Suppose that Mary utters (9a). Then, since the reference of the pronoun I is the speaker in a given context (see 9b), the referent of I in (9a) is Mary (see 9c), and therefore, (9a) is true if and only if Mary is rich. Similarly, if Tom utters (9a), then the reference of I in (9a) is Tom (see 9d), and (9a) is true if and only if Tom is rich. Unlike indexicals, the interpretation of definite descriptions does not depend on the utterance context but on the evaluation parameter, that is, the world of evaluation w. Consider (10).

(10) a. The president of GM must be rich. b. [[the president of GM]]c*,w

= [[the president of GM]],w = the person who is the president of GM in w

c. [[The president of GM must be rich]]c* is true if and only if, in every one of s*’s epistemic alternatives w', the president of GM in w' is also rich in w.

Here, unlike the speaker of the utterance, the denotation of the definite description the president of GM is not fixed in the utterance context c*, so its denotation varies depending on the world of evaluation, which is bound by the modal operator must in the relevant de dicto reading. Therefore, the de dicto reading of (10c) is true if and only if, in every epistemic alternative of the speaker s*, the president of GM is rich in that alternative, no matter who the president of GM is in that alternative. In this sense, the reportative evidential -ta- is also indexical. Consider (11).

(11)

[[John-i

Bill-ul

po-ass-ta-nta.]]c*

John-Nom

Bill-Acc

see-Past-ta-Decl

! 140

‘John saw Bill’ Presupposition: Someone told s* that John saw Bill.

Suppose that speaker s* utters (11) in c*. If s* in c* is Mary, then the implication introduced by ta- is that someone told Mary that John saw Bill. Similarly, if s* in c* is Tom, then the implication is that someone told Tom that John saw Bill. This shows that the reportative evidential implication of declaratives is anchored to the utterance speaker and, therefore, that non-final -tabehaves like an indexical. However, the parallel between -ta- and other indexicals is not observed in questions, as shown in (12a) and (12b), repeated from (1b) and (2b), respectively.

(12) a. John-i John-Nom

na-lul

po-ass-ta-nta.

I-Acc

see-Past-ta-Decl

‘John saw you.’ Presupposition: The speaker has heard from someone else that John saw the speaker. b. John-i John-Nom

na-lul

po-ass-ta-ni?

I-Acc

see-Past-ta-Q

‘Did John see you?’ Presupposition: The answerer has heard from someone else whether John saw the questioner.

Recall from the introduction that -ta2- and other indexicals behave differently in questions, in that while the evidential implication is shifted from the questioner’s perspective to the answerer’s perspective, the reference of an indexical pronoun in a question is always assigned from the questioner’s perspective. This is what I call the puzzle of the evidential perspective shift in ! 141

questions. Before turning to my solution for this puzzle, I illustrate Hamblin’s (1973) semantics of questions, which I adopt in what follows.

4.2.2. Hamblin (1973): questions as sets of answers

According to Hamblin (1973), the denotation of a question is the set of its instantial possible answers, that is, a set of propositions. Within this theory, the shift from propositions to sets thereof takes place at the level of the extension of wh-words. For example, a wh-word such as who is a set of individual concepts, the typical intensions of individual denoting expressions. Thus, the denotation of who is as in (13), which is type of <,t>.

(13)

[[who]]¢ = {%w. x: x'De}70

According to the type-driven semantics introduced by Heim and Kratzer (1998), one-place predicates are of type (and their intensions of type >), and two-place predicates are of type > (with intensions of type >>). Examples are given in (14).

(14) a. For any w, [[left]]w = %x. x left in w b. For any w, [[met]]w = %x. %y. y met x in w

([[left]]¢ = %w. %x. x leaves in w) ([[met]]¢ = %w.%x.%y. y met x in w)

Given all this, a special semantic rule, the so-called Pointwise Functional Application (PFA) rule (which is an amendment of Hamblin’s original compositional rule), is introduced in order to combine sets of individual concepts with ordinary predicates of type (or >) or !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 70

Following Heim and Kratzer (1998) and von Fintel and Heim (2007), I use the subscript ¢ to indicate the intension of a given extension.

! 142

properties of type > (or >>). This is merely a notational variant of the set tolerant combinatorial rule Hamblin proposes. The definition is given below.

Pointwise Functional Application Rule71

(15)

If + is a branching node and {,,-} is the set of its daughters, then: [[+]]¢

= %w. [[,]]¢(w) ([[-]]¢(w))

(i)

or

= {%w.[[,]]¢(w)(x(w)): x ' [[-]]¢}

(ii)

or

= {%w.f(w)([[-]]¢(w)): f ' [[,]]¢}

(iii)

or

= {%w.f(w)(x(w)): f'[[,]]¢ & x ' [[-]]¢}

(iv)

whichever is defined.

Finally, relative to the syntax of questions, Hamblin makes the following two assumptions: that questions contain no question morpheme and that wh-words are interpreted in their base position. Here is a concrete illustration of how Hamblin’s (1973) proposal works. The LF of a question such as Who left? is given in Figure 4.2.

Figure 4.2. LF of Who left? 3 who left Suppose that there are only three individuals in the domain: John, Bill, and Tom. Then the denotation of who is the set of individual concepts shown in (16).

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 71

This is a rendition of Hamblin’s (1973) original proposal made by Heim and von Fintel (2001). I quote this from Guerzoni (2010).

! 143

(16)

[[who]]¢

= {%w. x: x'De} = {%w. John, %w. Bill, %w. Tom}

From (14a) and (16), we derive the denotation of Who left? in the following way:

(17)

[[who left]]¢ (Each member in the set denoted by who is combined with the predicate left via 15ii) = {%w. [[left]]¢(w)(z(w)): z'[[who]]¢} = {%w.[%x. x left in w](z(w)): z'[[who]]¢} = {%w. z(w) left in w: z'[[who]]¢} = {%w. x left in w: x'De} = {%w. John left in w, %w. Bill left in w, %w. Tom left in w}

The resulting denotation for the question Who left? is the set of the three possible instantial answers to the question. Asking the question amounts to presenting these alternatives and asks the answerer to select from among them the one or more propositions he/she believes to be true. Turning now to yes/no questions, Hamblin (1973) assumes that they always contain a possibly covert whether with the following semantics:

(18)

[[whether]] = {%p.p, %p.~p}

This is a set that contains two functions from propositions to propositions, which maps their argument onto the affirmative and negative answers, respectively.

! 144

With this lexical entry, the denotation of the question did John leave? can be derived in the following way:

(19)

[[Did John leave?]]¢

[[whether]] = {%p.p, %p.~p} [[John left]] = %w. John left in w (The proposition that John left is combined with two functions in whether via PFA, therefore...)

[[Did John leave?]]¢

= {f(%w. John left in w): f'{%p.p, %p.~p}} = {[%p.p](%w. John left in w), [%p.~p](%w. John left in w)} = {%w. John left in w, %w. John did not leave in w}

Here the proposition that John left is combined with the denotation of whether via PFA, that is, each of the two functions in the denotation of whether apply to the proposition that John left in turn, and the two resulting values are collected in the set whose elements ultimately correspond to the affirmative and negative answers, respectively. As Garrett (2001) correctly observes, given the nature of the perspective shift, the parameter of the answerer would need to be recoverable at some point from the semantics of the question itself. In the following section, I show (contra Garrett) how this is possible without any substantial change to the semantics of interrogatives illustrated here. In fact, this approach to questions straightforwardly predicts that interrogatives can denote sets of objects different from propositions and still be suitable to represent possible instantial answers. In the case under consideration, these objects are characters, which serve to derive the evidential perspective shift in questions in a fully compositional fashion. The next section illustrates this point in detail.

! 145

4.3. A novel account for the evidential perspective shift in questions

This section illustrates my account of the evidential perspective shift in questions. The main ingredient of this account is the meaning I propose for evidentials. In the previous chapter, I claimed that Korean evidentials introduce an evidential presupposition. Based on these observations, I propose the following lexical entry for -ta-, where the evidential implication is introduced as a presupposition:

(20)

The lexical entry for -ta2For any utterance context c*, [[-ta2-]]c* = %pst.%c: sc has only reportative evidence relative to p. p

In other words:

(20')

For any context c* and c and proposition p, [[-ta2-]]w,c*(p)(c) is defined if and only if the speaker in c has reportative evidence only about p. If defined, [[-ta2-]]w,c*(p)(c) = p72

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 72

If the lexical entry for -ta- includes any epistemic modal component, and the apparent quantificational variability is accounted for in terms of a choice function (Matthewson et al. 2007), then the lexical entry for -ta- will be (20'), where f is a choice function provided with by the context, and the ordering source is omitted for simplicity: (20')

The lexical entry for -ta2- (modalized version) For any utterance context c*, [[-ta2-]]c* = !pst.!c: Bc = {p: sc believes that p is true only based on the reportative evidence}. !w. &w'[w''f(Bc(w))*p(w')] (c is a variable over contexts, that is, the triple , and f is a choice function of type )

As pointed out in the previous chapter, since at the moment there is no compelling evidence that Korean evidentials contain epistemic modality as its assertive component, and furthermore, since epistemic modality is not relevant to the evidential perspective shift, I will use the simple lexical entry for -ta- as in (20), rather than in (20'). Note, however,

! 146

According to this proposal, -ta- is a function from propositions to partial characters, introducing the definedness condition that the speaker sc in the context the character applies to has reportative evidence that p is true. Notice that this presupposition is essentially an expressive presupposition in the sense of Schlenker (2007): it is anchored to the context, may introduce new information that does not exist in the conversational background (unlike pragmatic presuppositions), and the addressee can easily accommodate it. When defined, and applied to a context for getting an intension, -ta- returns the same proposition it applies to. My proposal makes two important predictions. First, if the resulting character is immediately applied to the utterance context, then the evidential presupposition relates to the utterance context c* and it is anchored to the speaker. This is what happens in declaratives, where the top node indeed denotes a character as described above. This is why -ta2- in declaratives shows the parallelism with indexicals noted above: both evidentials and indexicals end up being interpreted relative to c*, that is, to the utterer. The second prediction concerns interrogatives. According to Hamblin (1973), asking a question amounts to presenting to the addressee a set of alternative propositions and requesting that the addressee pick up those propositions in the set that he/she believes to be true and utter them. Given this, none of the answers in this set is directly uttered by the questioner. However, any answer is directly uttered only when the addressee chooses it as a true answer. This clearly holds regardless of whether we conceive those answers as propositions or as characters. Now, if a question is a set of characters, each of which carries the reportative evidential presupposition introduced by -ta-, then the presupposition in each answer is still anchored to the context c, which is bound in the character itself, and therefore, because the questioner does not utter any of these !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! that my lexical entry for -ta- in (20) can be integrated with other proposals which analyze evidentials as epistemic modals (Izvorski 1997, Matthewson et al. 2007, among others) without any problem, as the lexical entry in (20') shows.

! 147

characters, this presuppositions will fail to be anchored to the questioner context. Only when the addressee answers the question are one or more of these characters (the correct answers) uttered, and, therefore, applied to the answerer’s context, hence the shift. Since the context parameter which is relevant for the interpretation of other indexicals is instead not bound by the character and, therefore, ends up being identified to the questioner’s context, the difference between indexicals and the evidential presupposition follow.73 In the rest of this section, I illustrate this point by applying the above proposal to two concrete examples.

4.3.1. Declaratives

Let us begin by considering the case of -ta2- in declarative sentences.

(21)

John-i

na-lul

po-ass-ta-nta.

John-Nom

I-Acc

see-Past-ta-Decl

‘John saw me’ Implication: The speaker has reportative evidence that John saw the speaker.

Since -ta2- takes sentential scope, the LF of (21) is as in Figure 4.3.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 73

The reader might want to recall the distinction between asserter and utterer given in Chapter 1: according to this distinction, evidentials are always anchored to asserter, whereas other indexicals are always anchored to utterer.

! 148

Figure 4.3. LF of (21) John-i na-lul po-ass-ta-nta. (ii) 3 -ta23 John 3 saw me (i) For any utterance context c* (i.e., the triple ), the denotation of (21) is derived as follows:

(22)

(i) [[me]]c* = s*. (ii) [[(21)]]c* = [[-ta2-]]c*([[John saw me]]c*)

(by Intensional Functional Application)

= [[-ta2-]]c*(%w. [[saw]](John)(s*))

(by i)

= [[-ta2-]]c*(%w. John saw s* in w)

(by the lexical entry for saw and %-reduction)

= %c: sc has reportative evidence that John saw s*. %w. John saw s* in w

An utterance of this character in the context c* amounts to its application to c*, as shown in (23a).74 The resulting intension is (23b).

(23)

[[(22)]](c*) a. = [%c: sc has reportative evidence that John saw s*. %w. John saw s* in w](c*) is defined if and only if sc has reportative evidence that John saw s*. b. If defined, [[(22)]](c*) = %w. John saw s* in w

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 74

But see Chapter 5 for a more detailed account of this claim: specifically, how both evidentials and indexicals end up being anchored to s* in c*.

! 149

More informally, an utterance of (21) in c* has a defined intension if and only if the utterance speaker s* has reportative evidence concerning whether John saw the speaker, and if defined, it has the proposition that John saw the speaker s* as its intension. Notice that both pronouns and evidential presuppositions are ultimately anchored to the utterance parameters of c*, as desired.

4.3.2. Questions

We can now turn to the case of -ta2- in questions such as (24).

(24)

John-i

na-lul

po-ass-ta-ni?

John-Nom

I-Acc

see-Past-ta-Q

‘Did John see me?’ Implication: The addressee is expected to have reportative evidence about whether John saw the speaker.

Recall from the previous section that according to the theory adopted here, a yes/no question contains a covert whether (Hamblin 1973). Given this, there are two logically possible scope options for whether and -ta2-, each of which I consider in turn. In Figure 4.4, -ta2- takes narrow scope with respect to whether.

Figure 4.4. whether takes wide scope over -ta3(i) whether 3 -ta23 John 3 saw me ! 150

The LF in Figure 4.4 is uninterpretable because, as we saw above, once -ta- applies to the proposition John saw the speaker, it returns the character in (25), a semantic object whether cannot semantically combine with (the lexical entry for whether is repeated for convenience).

(25)

[[(i)]]c* = %c: sc has reportative evidence that John saw s*. %w. John saw s* in w

(26)

[[whether]]c* = {%p.p, %p. ~p}

Recall from the previous section that a semantic object such as the one in (26) combines with the denotation of its sister node by PFA. The rule essentially applies each function in the denotation of whether to the denotation of its sister node and then collects the values of each application in a set. Given this, and because each function in the set denoted by whether is of type , whether can combine only with a proposition (of type ) and not with a character (of type ). The other possible LF for (24) is given in Figure 4.5, where -ta2- takes scope over the entire question whether John saw the speaker in c*.

Figure 4.5. LF of (24) John-i na-lul po-ass-ta-ni? (iv)<,t> 3 (iii) -ta-> 3 (ii)t->st whether<,t> 3 et Johne 3 saw mee (i) This option is interpretable and generates the desired denotation for the question. (27) provides a detailed illustration of how this denotation is derived. ! 151

(27)

For any context c* () (i) [[me]]c* = s* (ii) [[John saw me]]c* = %w. John saw s* in w (iii) (two functions in whether is applied to the proposition that John saw me via PFA) -> {f(%w. John saw s* in w): f'{%p.p, %p.~p}} = {[%p.p](%w. John saw s* in w), [%p.~p](%w. John did not see s* in w)} = {%w. John saw s* in w, %w. John did not see s* in w} (iv) (-ta- is applied to two propositions in iii via PFA) -> {[[-ta-]]c*(p): p'[[whether John saw me]]c*} = {[[-ta-]]c*(%w. John saw s* in w), [[-ta-]]c*(%w. John did not see s* in w)} = {%c: sc has reportative evidence that John saw s*. %w. John saw s* in w, %c: sc has reportative evidence that John did not see s*. %w. John didn't see s* in w}75

As the above computation indicates, we predict that an interrogative containing -ta2- will always denote a set of characters. Unlike in the case of declaratives, however, none of the characters binding the evidential presupposition is directly uttered by the speaker of the question. Rather, the whole set is presented to the addressee by the questioner as is. It then follows that, technically, none of the characters in the question denotation ends up being applied to the questioner’s context. The addressee alone is the utterer of his/her choice of characters in that set (the true answers), and therefore, the evidential presupposition ultimately applies to him/her. This is exactly the desired result. The evidential in the question obligatorily undergoes a perspective shift from the perspective of the questioner to that of the answerer. The shift is fully predicted based on the way !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 75

In the negative answer in (27), since the argument of -ta2- is the negation of the (ii), the evidential presupposition is also changed accordingly.

! 152

I proposed encoding evidentiality in the lexical contribution of particles such as -ta- and on the semantics of questions in general. To summarize, in this section, I proposed that the evidential morpheme is a function from propositions to characters and that the implication it introduces is an expressive presupposition, encoded in terms of a definedness condition on contexts directly. It follows from my analysis that -ta- takes wide scope in questions and that questions containing -ta- denote sets of characters rather than sets of propositions. I have also shown how the latter is only a minor departure from Hamblin (1973)’s theory, being fully predicted once functions returning characters are allowed in the system. Given this, without any fundamental revision of Hamblin’s (1973) semantics of questions, the present account for the evidential perspective shift in questions is merely based on the semantics of the reportative evidential marker -ta2- and its interaction with the semantics and pragmatics of questions.

4.4. A plausible (but incorrect) alterative: the pragmatic account

Under the account I presented in the previous section, the evidential perspective shift emerges as a fully semantic phenomenon, but it is reasonable to wonder whether it should instead be viewed as a pragmatic effect. Although no fully explicit pragmatic account exists, other scholars have pointed in this direction (e.g., Garrett 2001). In this section, I attempt to make concrete sense of what the pragmatic motivation for a shift in evidential perspective in questions could most plausibly be. I then argue that, insofar as evidentiality in Korean is concerned, a pragmatically based account in the terms outlined in this section would be empirically unsatisfactory.

! 153

When a questioner Q is asking a question about an issue P76 to an answerer A under genuine information-seeking circumstances, we can probably assume that Q has no opinion about the truth or falsity of P. One might argue that this state of ignorance concerning Q should correspond, under most circumstances, to the unavailability to Q of any type of evidence concerning whether P is true or false, in order for Q to be entitled to ask a question about P. If correct, this would explain why evidentials in questions do not generally refer to the evidence available to the questioner and why, since the addressee’s information state is the only other salient information state in the discourse, evidentiality regards it instead. This would also apply to indirect evidentials. In terms of evidence, when a conversation is felicitous, Q expects A to answer based on the most informative evidence. However, suppose that Q expects A not to have any direct evidence (which is more informative than reportative evidence), but to have only reportative evidence regarding P. Then Q may indicate this expectation about A by employing the reportative evidential marker when asking a question. In this case, the reportative evidential implication is shifted from questioner to addressee in questions. However, since the explanation sketched above is based on a pragmatic notion of what information state of the speaker makes it plausible for him/her to ask a genuine question, it does not exclude the possibility of anchoring evidence to the speaker in questions on principled terms. Whenever this option is compatible with the plausibility of asking a question, it should become available and the shift should become unnecessary. Here is an example of a context where we expect exactly this to happen. If Q only has reportative evidence but expects A to have stronger evidence (i.e., direct evidence) regarding P, then for Q to ask A about P is a reasonable conversational move in which Q can still express his/her own indirect evidentiality, thus justifying his questioning and expecting the addressee to be better informed. A typical case of this !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 76

I use the term issue without commitment to any theory.

! 154

sort is the interrogation of a suspect in the course of an investigation. Suppose a detective Q is at a stage of investigating a case P where he has only collected clues based on reportative evidence. Based on these clues, Q suspects A to be the culprit and, thus, (obviously) to have direct evidence regarding P. Q can attempt to verify the clues by interrogating A. According to the pragmatic view sketched above, in this context, Q’s questions containing reportative evidentials should be acceptable. As the following two examples clearly indicate, this prediction is not borne out. First consider (28).

(28)

(Scenario: John is a detective and is investigating a case of embezzlement in a company. While interviewing several employees of the company, John hears that Mary is the one who embezzled the money. Finally, he interviews Mary and asks...) #Tangsin-i you-Nom

ton-ul

hoynglyengha-ess-ta-pnikka?

money-Acc

embezzle-Past-ta-Q

‘Did you embezzle the money?’ Implication (that would fit the context): I only have indirect evidence about whether you embezzled the money. Did you do so? Actual Implication: The addressee has reportative evidence about who embezzled the money.

In this case, the questioner has reportative evidence that Mary is the embezzler, and the addressee is expected to have direct evidence of this (regardless of whether the true answer to the question is affirmative or negative). In this context, only a speaker-oriented reading of the reportative evidential would be appropriate. Therefore, if the evidential shift for -ta- were to be understood in the pragmatic terms outlined in this section, -ta- should be felicitous in this question, and the reading should be a non-shifted one. The fact that -ta2- is instead infelicitous is a clear indication ! 155

that a non-shifted reading is not available. We can therefore conclude that -ta2- is obligatorily related to the answerer rather than the questioner It is unnatural for John to expect Mary to have any reportative evidence about what she has done, since he is asking a question to a person who he suspects is the embezzler. (29) leads us to the same conclusion.

(29)

(Scenario: John is a detective and is investigating a case of embezzlement in a company. From one of the employees in the company, John heard that Mary is the one who embezzled the money. He also found that another employee, Tom, was in the office at the same time when Mary is suspected of having embezzled the money. John asks Tom...) a. #nwu-ka Who-Nom

ton-ul

hoynglyengha-ess-ta-pnikka?

money-Acc

embezzle-Past-ta-Q

‘Who embezzled the money?’ Implication: the addressee has reportative evidence about who embezzled the money. b. #Tangsin-i you-Nom

ton-ul

hoynglyengha-ess-ta-pnikka?

money-Acc

embezzle-Past-ta-Q

Implication: the addressee has reportative evidence about who embezzled the money.

Here again, the speaker expects the addressee to have direct (rather than reportative) evidence that Mary is the embezzler. If the evidential perspective shift triggered by -ta2- in questions were optional, (29a) and (29b) would be felicitous, contrary to the facts. To summarize the discussion above, both (28) and (29) show that the evidential perspective shift triggered by -ta2- in questions is obligatory rather than optional. This strongly suggests that the perspective shift stems from the semantics of the evidential, as proposed in the previous section, rather than from pragmatic factors related to questioning and information states. ! 156

4.5. Review of previous accounts

To the best of my knowledge, the proposal I made in Section 4.3. is the only existing analysis of the evidential perspective shift in Korean. However, similar evidential shifts in questions are attested in other languages and have been discussed in literature on evidentiality. In the following two subsections, I briefly illustrate the two predominant positions on this issue and argue that they cannot be extended to -ta- or other Korean evidentials. Moreover, I discuss other speaker-anchored expressions that also appear to trigger perspective shifts in questions and show that these effects are also substantially different from the perspective shift in Korean evidentials.

4.5.1. Garrett (2001): Questions as sets of assertions

Garrett (2001) observes that Tibetan questions containing indirect evidentials undergo perspective shifts of the sort I am discussing in this thesis. Garrett (2001) proposes that a perspective shift can only be explained if the extension of a question contains reference to both the speaker’s and addressee’s parameters, which leads him to modify the semantics of interrogatives to make their extensions sets of assertions (technically “utterances”) rather than sets of propositions. Specifically, according to Garrett, the extension of a question such as Who left? should be (31) rather than (30).

(30)

The extension of Who left? in Hamblin (1973) (quoted in Garrett 2001: 237) Answer-set(Who left?) = {p|!x. p = x left}

! 157

In words: the answer set for the question Who left? is the set of propositions p such that there is an x such that p = x left.

(31)

Revised extension of Who left? (Garrett 2001: 237) Answer-set(Who left?) = {A|!x. Content(A) = x left} In words: the answer set for Who left? is the set of assertions A from h to s such that there is an x such that the content of A is that x left.

Given this semantics for questions, Garrett (2001) suggests that the choice between a shifted and a non shifted reading is pragmatically driven in a way that is similar to the proposal I discussed above.

Origo shift [corresponding to what we call the evidential perspective shift: DL] is a mere pragmatic necessity. Because assertions have authors and (sometimes also) recipients, an assertion must have an origo. The origo is simply the author of the assertion, which is not determined grammatically, but rather pragmatically. For example, if I ask someone a question, the answer-set consists of assertions by the person I’m talking to, i.e. the person right there in front of me, simply because that’s the person who is supposed to answer my question. No formal machinery is required: the origo shift comes for free as long as assertions have authors and questions have answerers. (Garrett 2001: 238)

Even though Garrett’s (2001) proposal is very similar to what I suggested above in the sense that we need the denotation of the question to be such that the addressee parameter is ultimately relevant to the intension of the possible answers, Garrett’s proposal cannot account for the evidential perspective shift in Korean for the following reasons. First, it is unclear whether the semantics Garrett suggests, paired with the pragmatic reasoning in the quote, predicts that the shift is obligatory or optional. We saw, however, that the evidential perspective shift is definitely ! 158

obligatory with Korean -ta- (recall the examples in the previous section). Furthermore, if the answer-set is a set of assertions where the questioner and the answerer are both specified, then each alternative (i.e., each ‘assertion’) in a given set, regardless of whether it contains evidentials or indexicals, may be evaluated with respect to the addressee rather than the speaker. In other words, under Garrett’s analysis, both the evidential perspective and other indexicals may be shifted. However, in Korean, the reference of indexicals does not shift in questions, contrary to Garrett’s (2001) prediction. Finally, and more generally, it is unclear whether the denotation of interrogatives Garrett suggests can be derived within a standard compositional view of the semantics of questions, a problem that, as we saw above, my proposal does not present.

4.5.2. Faller (2002): Cuzco Quechua reportative evidential -si- in questions

Discussing evidentials in Cuzco Quechua, Faller (2002) reports that the question in (32) is ambiguous between the readings in (32i) and in (32ii). (32ii) is similar to what we have observed for Korean questions with -ta-, whereas in (32i), the semantic contribution of -siappears to be just like that of a quotation predicate such as ‘say.’

(32)

Pi-ta-s

Inés-qa

watuku-sqa?

who-acc-si

Inés-top

visit-pst2

‘Who did Inés visit?’ (i) The speaker indicates that somebody else is asking (ii) The speaker expects the hearer to have indirect evidence for his/her answer (Faller 2002: 230, ex. 189b)

! 159

Like the Cuzco Quechua reportative evidential -si-, the Cuzco Quechua evidential -mi-, indicating that the speaker’s assertion is based on his/her best possible grounds (hereafter BPG) is also ambiguous in questions.

(33)

Pi-ta-n

Inés-qa

watuku-sqa?

who-acc-mi

Inés-top

visit-pst2

‘Who did Inés visit?’ (i) The speaker has the best possible grounds for asking (ii) The speaker expects the hearer to base his/her answer on the best possible grounds (Faller 2002: 230, ex. 189a)

For convenience, let us refer to (32i) and (33i) as the questioner-anchored reading, and (32ii) and (33ii) as the addressee-anchored reading, the latter being the shifted reading I am discussing in this chapter. To explain Faller’s (2002) analysis, let us begin with the lexical entries that Faller provides for the Cuzco Quechua BPG evidential -mi- and reportative evidential -si-. As we saw in Chapter 1, according to Faller, mi and si are functions from speech acts to speech acts.

(34) a. mi:

ILL

(v)

SINC

b. si:

ILL

= {M(s,v)}

(v)

SINC

->

= {M(s,v)}

ILL

(v)

SINC

->

= {M(s,v), Bpg(s, M(s,v))}

PRESENT SINC

(v)

= {!s2[Ill(s2,p) " s2 # {h,s}]} (Faller 2002: 231, ex. 190)

! 160

In (34), ILL is a variable over illocutionary force markers, Ill is the predicate corresponding to the illocutionary force ILL, M is a predicate over propositional attitudes such as Belief (Bel) or Desire (Des), and v is a variable over entities that can serve as the second argument of a mental attitude such as a proposition or a speech act. If these two functions are applied to an assertion of a proposition p, then the argument and the outcome of mi and si, respectively, are as shown in (35):

(35) a. mi:

ASSERT SINC

b. si:

->

= {Bel(s,p)}

ASSERT SINC

(p)

(p)

ASSERT SINC

->

= {Bel(s,p)}

(p)

= {Bel(s,p), Bpg(s, Bel(s,p))}

PRESENT SINC

(p)

= {!s2[Assert(s2,p) " s2 # {h,s}]} (Faller 2002: 231, ex.190)

According to (35), mi changes the sincerity condition in requiring that the speaker believe that his/her assertion is based on the best possible ground. si changes the sincerity condition in requiring that there be another individual, who is neither the speaker nor the addressee, who asserts p. In addition to this, si changes the speech act from assertion to presentation, indicating that the speaker does not make any commitment to the truth or falsity of the proposition p. Faller further assumes that to ask a question is to request the hearer’s assertion of a proposition. Therefore, the speech act QUEST(ION) is the speech act REQUEST, which takes ASSERT(ION) of q as its argument.

(36)

QUEST = REQUEST (ASSERTh (q)) (Faller 2002: 237)

! 161

Since a question is a request, the sincerity condition for a question is as in (37), where Des (x, ay) indicates that x desires y to perform a, Fx(p) indicates that x performs the speech act F(p), and s and h refer to the speaker and hearer, respectively.

(37)

SINC

= {Des(s, Fh (p))}

Given this, Faller argues that the ambiguity between the questioner-anchored reading and the addressee-anchored reading can be accounted for in terms of a peculiar scope ambiguity of illocutionary operators, that is, the illocutionary speech act EVI (evidence) may take wide scope over REQUEST or narrow scope with respect to REQUEST, resulting in two different ianchoring of evidentiality. Specifically, she claims that, first, the questioner-anchored reading follows if the evidential takes widest scope as in (38).

(38)

Reading (i):

EVI (REQUEST (ASSERTh (q))) = EVI (QUEST)

(questioner-anchored) (Faller 2002: 237)

Given this, the questioner-anchored reading can be accounted for by applying (35) to a question whose sincerity condition is as in (37). The functions denoted by mi and si, when they apply to questions, are shown (39).

(39) a. mi:

QUEST SINC

b. si:

->

= {Des(s, Fh(p))}

QUEST SINC

(p)

(p)

= {Des(s, Fh(p))}

QUEST SINC

->

(p)

= {Des(s,Fh(p)), Bpg(s, Des(s, Fh(p)))}

PRESENT SINC

(p)

= {!s2[Quest(s2,p) " s2 # {h,s}]} (Faller 2002: 231, ex. 190) ! 162

Secondly, according to Faller, the addressee-anchored reading can be accounted for by simply assuming that the illocutionary operator REQUEST takes wide scope over the evidentials.

(40)

Reading (ii):

REQUEST (EVI (ASSERTh (q)))

(addressee-anchored) (Faller 2002: 237)

Faller does not give any more detail concerning her analysis of the addressee-anchored reading of mi and si, but here let us follow how Faller’s formalization of Cuzco Quechua evidentials would derive reading (ii) in (40). For brevity I will focus on si, the lexical entry for which in Faller (2002) is repeated below for convenience:

(41)

si:

ILL

(v)

SINC

->

= {M(s,v)}

PRESENT SINC

(v)

= {!s2[Ill(s2,p) " s2 # {h,s}]} (Faller 2002: 231, ex. 190)

According to (40), si first applies to an assertion and returns a speech act of presentation, just like when it appears in declaratives, as illustrated above (repeated below as 42):

(42)

si:

ASSERT SINC

(p)

= {Bel(s,p)}

->

PRESENT SINC

(p)

= {!s2[Assert(s2,p) " s2 # {h,s}]} (Faller 2002: 231, ex. 190)

As we saw above, for Faller, asking a question ends up to requesting a speech act, and the sincerity condition for question is as below (repeated from above): ! 163

(43)

SINC

= {Des(s, Fh (p))}

However, when REQUEST takes wide scope over EVI as in (40), it is unclear how this sincerity condition is applied to the result of the application of si to the speech act PRESENTATION. The best way to do that, as far as I can see, is to conjoin two sincerity conditions (one from question, and the other from presentation). Then the result of applying the speech act QUESTION to the result of (42) is as (44). Specifically, the left hand side is the result of (44), where si is applied to the assertion, and the right hand side is the result of applying QUESTION to the speech act PRESENTATION

(44)

introduced by -si-:

PRESENT SINC

(v)

->

= {!s2[Ill(s2,p) " s2 # {h,s}]}

QUESTION SINC

(v)

= {Des (s, PRESENTh (p)) & !s2[ASSERT(s2,p) " s2 # {h,s}]}

The sincerity condition in (44) is read as follows: a speaker requires a hearer to present a proposition asserted by another speaker s2, who is neither the speaker nor the addressee. I am not sure whether this sincerity condition is what we expect from Reading 2: it seems to me that in Reading 2, a speaker is requesting a hearer to assert based on what the hearer heard from someone else, which is different from what the sincerity condition in (44) is saying. Furthermore, it is rather clear that, if it achieves the purpose Faller states it for, then her view cannot be correct for Korean, since -ta- in questions does not allow the reading in (45i).

(45)

Inés-nun

nwukwu-lul

manna-ess-ta-ni?

Inés-top

who-Acc

meet-Past-ta-Q ! 164

‘Who did Inés meet?’ (i) The Speaker indicates that somebody else is asking

=> does not exist

(ii) The speaker expects the addressee to have indirect evidence for his/her answer

Another potential problem for a view that appeals to illocutionary operators is that it might obscure the similarities and differences between indexicals and evidentials discussed in this chapter. In addition to all of these problems we saw above, in Faller’s account of the questioneranchored reading of mi, it is unclear to me what it means that the speaker has the best possible grounds for asking. Faller herself agrees that -mi in questions is much less ambiguous than -si in questions (judgments with -mi- in questions seem much more vague).77 According to Faller, the speaker-anchored meaning of mi in questions could mean two things: “(i) the speaker has authority over the hearer and therefore has the best possible grounds for demanding an answer, or (ii) the speaker has very good reasons for wanting to know the answer - or both” (Faller 2002: 232). An example for (i) is given in (46a) asked by a teacher, and another example illustrating (ii) is given in (46b) asked by a customer.

(46) a. Hayk’a-n

iskay

how-much-mi two

yapa-sqa

iskay-man?

add-PP

to-Illa

‘How much is two plus two?’

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 77

If -mi in questions turns out not to be ambiguous after all, the ambiguity of questions with -si- might find an alternative explanation in the possibility that si itself is lexically ambiguous between a true evidential and a quotational operator, just like -ta- and -ay (the phonological contraction of the pro-verb) in Korean, respectively as we saw in Chapter 2.

! 165

b. Hayk’a-n

vale-n chay?

How-much-mi cost-3 this ‘How much does this cost?’

However, in my view, the problem is that in both examples the addressee-anchored reading of mi is equally available. (46a), as well as (46b), are compatible with the addressee-anchored implication that the speaker is expecting the addressee to answer the question based on his/her best possible ground. Faller also agrees that there are ambiguities in (some) questions with mi between the speaker-anchored reading and the addressee-anchored reading, but she does not provide us with any more data where this ambiguity is resolved. Summarizing the discussion above, we saw that previous approaches to phenomena similar to those discussed in this chapter fail to extend to the evidential perspective shift in questions triggered by -ta-. Garrett’s (2001) analysis does not account for the obligatory shift of ta2- in questions and the lack of shift in indexicals. Furthermore, unlike Cuzco Quechua evidentials in Faller (2002), the reportative evidential -ta2- cannot be an illocutionary operator with the potential to scope outside the illocutionary operator REQUEST.

4.6. Evidential -ta- vs. other perspective-anchored components

Several expressions in natural language have interpretations that vary depending on the speaker, including expressives (e.g., Potts 2005, Amaral et al. 2007, Potts 2007, Harris & Potts 2010) and utterance-modifying adverbs (Garrett 2001; speaker-oriented adverbs in, e.g., Ernst 2009). This section discusses their behavior in questions. As we will see below, the perspective shift in questions triggered by expressives and utterance-modifying adverbs is different from the ! 166

evidential perspective shift triggered by -ta2- in that it is optional. This leads to the conclusion that the evidential implication introduced by -ta2- must be different from that introduced by other perspective-dependent expressions.78

4.6.1. Expressives: Optional shift

Potts (2005) claims that the interpretation of expressives is always speaker-dependent (i.e., the perspective of expressives is always the speaker’s), but Amaral et al. (2007) and Harris and Potts (2010) provide apparent counterexamples to this claim. One such example appears (47), where the expressive idiotic is not speaker-dependent, but dependent on the perspective of everybody in the room, given the fact that Lewis Lapham is a liberal supporting civil liberties and constitutional rights.

(47)

I was struck by the willingness of almost everybody in the room — the senators as eagerly as the witnesses — to exchange their civil liberties for an illusory state of perfect security. They seemed to think that democracy was just a fancy word for corporate capitalism, and that the society would be a lot better off if it stopped its futile and unremunerative dithering about constitutional rights. Why humor people, especially poor people, by listening to their idiotic theories of social justice? [Lewis Lapham, Harper’s Magazine, July 1995] (from Harris & Potts 2010: 7)

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 78

Potts (2005) claims that expressives and utterance-modifying adverbs introduce conventional implicatures, but Potts (2007), discussing the semantic contribution of expressives, simply adopts the term expressive content (though he still adopts the formalization of conventional implicature in Potts 2005 to analyze expressive content).

! 167

Similarly, in questions (and possibly in declaratives), the perspective of expressives may or may not shift from the speaker’s to the addressee’s in a way that is reminiscent of the facts discussed in this thesis. Consider (48), where the implication introduced by pilemekul ‘damn’ can be interpreted in two ways: the speaker may think that the addressee’s job is quite bad, or the addressee may think that his/her own job is quite bad.

(48)

ne-nun

ku pilemekul il-ul

kumantwu-ess-ni?

You-Top

that damn job-Acc

quit-Past-Q

‘Did you quit that damn job?’

However, as noted was above, the shift in (48) is not obligatory and is probably pragmatically determined.

4.6.2. Utterance-modifying adverbs: optional shift

In questions, the utterance-modifying adverb solcikhi ‘honestly’ is also ambiguous with respect to whether it is anchored to the speaker or to the addressee (cf. Garrett 2001 and Potts 2005).

(49)

Solcikhi,

John-i

ne-lul

manna-ess-ni?

Honestly

John-Nom

you-Acc

meet-Past-Q

‘Honestly, did John meet you?’

! 168

The sentential adverb solcikhi ‘honestly’ in (49) can be either interpreted from the addressee’s perspective or the speaker’s. When interpreted from the addressee’s perspective, (49) carries the implication that the speaker asks the addressee to answer in an honest way. When interpreted from the speaker’s perspective, however, (49) carries the implication that the speaker’s curiosity in asking the question is genuine (or that he/she really does not know anything about the answer). This indicates that in this case, the perspective shift with utterance-modifying adverbs in Korean is optional, and pragmatic factors might very well determine whether the shift occurs, unlike in cases of evidentials.79 To conclude, I have shown that the presuppositions introduced by -ta- differ from the implications introduced by expressives and utterance-modifying adverbs, in that the perspective shift triggered by expressives and utterance-modifying adverbs in questions is optional. The general question of how to explain optional shifts of this sort remains open.

4.7. Korean direct evidential -te-

Before closing this chapter, let me briefly discuss the Korean direct evidential -te-. At first, one might think that my proposal could easily be extended to the direct evidential -te-. For example, -te- also shows the perspective shift in questions, as shown in (50) and (51). !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 79

Andrew Simpson (p.c.) points out that when honestly appears to be anchored to the speaker, the question carries the implication that the speaker is negatively bised to the answer of the question. Consider the following question, which is the translation of (39): (i)

Honestly, did John meet you?

When honestly is anchored to the addressee, as we saw above, the question carries the implication that the speaker asks the addressee to answer honestly (without lying). However, when anchored to the speaker, in addition to the implication we saw above, it also seems to carry the implication that the speaker does not believe that John met the addressee. Like English, (49) also seems to carry the same implication when the sentential adverb solcikhi ‘honestly’ is anchored to the speaker. Given this, it is plausible to assume that sentential adverbs like solcikhi/honestly are always addressee-anchored in questions, and the apparent speaker-anchored reading of these adverbs is only a pragmatic byproduct. In the next chapter I will discuss more in detail how this new proposal can account for the speaker-anchored reading of sentential adverbs, and what the implication of this second proposal might be.

! 169

(50)

John-i

Mary-lul

manna-te-la.

John-Nom

Mary-Acc

meet-te-Decl

‘John met Mary.’ Implication: The speaker has direct evidence that John met Mary.

(51)

John-i

Mary-lul

manna-te-nya?

John-Nom

Mary-Acc

meet-te-Q

‘Did John meet Mary?’ Implication: The addressee is expected to have direct evidence about whether John met Mary.

One might think that this could also be accounted for by assuming that -te- is a function from propositions to characters.

(52)

The lexical entry for -te- (first attempt) For any utterance context c*, [[-te-]]w,c*(p)(c) = %pst.%c: sc has only direct evidence relative to p. %w. p(w)=1 (where c is a variable over contexts, i.e., a triple of )

Some might object to this proposal by pointing out that this lexical entry overlooks several peculiar aspects of -te-, among which I would like to discuss two in this conclusion. Actually, they are the two aspects that I pointed out in previous chapters, and for convenience I repeat these below. First, as many Korean grammarians have noted, -te- seems to encode a past tense meaning. Consider the following example, which I again repeat from Chapter 2 for convenience: ! 170

(53)

John-i

ecey

Mary-lul

manna-te-la.

John-Nom

yesterday

Mary-Acc

meet-te-Decl

‘John met Mary yesterday.’ Implication: the speaker has direct evidence (e.g., the speaker saw) that John met Mary.

Without any tense marker, (53) seems to indicate that the prejacent denotes an eventuality that occurred before the time of utterance, as indicated by the temporal adverbial ecey ‘yesterday’. The second problem is that -te- seems to introduce evidential implications different from direct evidentiality when it combines with other tense/aspect/mood morphemes. Specifically, in such cases, -te- seems to introduce the inferential evidential implication. Consider the following examples, which I also repeat from Chapter 2 for convenience:

(54) a. (Scenario: yesterday I saw that John met Mary) John-i

ecey

Mary-lul

John

yesterday

Mary-Acc

manna-te-la. meet-te-Decl

‘John met Mary yesterday.’ Implication: the speaker has direct evidence that John met Mary. b. (Scenario: yesterday I found that John had Mary’s book, which I had seen before was in Mary’s home) ?John-i

kucey

Mary-lul

manna-ess-te-la.

John-Nom

the-day-before-yesterday

Mary-Acc

meet-Past-te-Decl

‘John met Mary the day before yesterday.’ Implication: based on the speaker’s inference, John met Mary the day before yesterday.

! 171

c. (Scenario: John has some problem with his homework and needs some help as soon as possible. I think Mary is the only one who can help him, but I saw Mary’s schedule where she is out of town today and will be back tomorrow.) John-i

nayil-i-lato

Mary-lul

manna-keyss-te-la.

John-Nom

tomorrow-L-at least/even

Mary-Acc

meet-may-te-Decl

‘John will meet Mary tomorrow at the latest.’ Implication: based on the speaker’s inference, John will meet Mary tomorrow.

It is obviously beyond the scope of this chapter (and perhaps beyond the scope of this thesis) to discuss every aspect of -te-. Therefore, let me briefly review previous proposals regarding -teand the type of problems these proposals have. This will help me to make another proposal for the lexical entry for -te- which might be a tentative solution to the problems discussed in this section. First, it is not always clear whether -te- actually encodes a past tense meaning. In Chapter 2, we saw that a sentence with -te-, without any other tense/modal marker, is also compatible with an adverbial anchored to the time of the utterance, corresponding to the English now.

(55)

cikum

pakk-ey-nun

pi-ka

o-te-la.

Now

outside-Loc-Top

rain-Nom

come-te-Decl

‘[I noticed] It is raining outside now.’ (Chung K.-S. 2007: 21)

As I discussed in Chapter 2, to account for such examples, Chung K.-S. (2007) proposes that -teis a deictic past tense marker rather than a direct evidential. Here is the lexical entry for -te- given in Chung K.-S. (2007), which I repeat from Chapter 1:

! 172

(56)

Spatial deictic past tense -te[[-te-]]c is only defined if c provides a unique salient spatiotemporal location Lc such that $(Lc) < tc " Lc % P-trace(sc). If defined, then [[-te-]]c = Lc. (where tc = the speech time, sc = the speaker of the context c) (Chung 2007: 204, ex. 26b)

Under this analysis, -te- encodes not only the temporal meaning (i.e., that of past tense) but also the spatial meaning. Thus, according to this lexical entry, very roughly speaking, the time of the eventuality denoted by the prejacent should be before the time of the utterance, and the place of the eventuality denoted by the prejacent should be different from the place of the utterance. However, we have already pointed out several problems that Chung’s proposal may have. First of all, to account for the inferential evidentiality in (54b) and (54c), Chung needs to assume the lexical ambiguity of the past tense marker -ess- and the modal marker -keyss- between their original meanings and the evidential meaning.

(57)

(when there is no deictic tense) Ø:

present tense

-ess-:

perfective

-keyss-:

mood

! 173

(58)

(when there is an overt deictic tense) Ø:

direct evidential

-ess-:

indirect evidential based on result states

-keyss-:

indirect evidential based on reasoning (Chung K.-S. 2007: 205)

Again, as I pointed out in Chapter 2 and as pointed out in Lee J. (2010), this is obviously ad hoc. Second, to argue for the deictic-tense analysis of -te-, Chung compares (55) with (59).

(59)

#cikum

yeki-eyse-nun pi-ka

o-te-la.

Now

here-Loc-Top

come-te-Decl

rain-Nom

(Intended): ‘[I noticed that] It is raining here now.’

However, we also have this type of unacceptability in epistemic modality. For example, a speaker cannot utter (60) while watching it rain outside.

(60)

It must be raining outside.

Thus, the parallelism between (59) and (60) tells us that the unacceptability of (59) does not need to be explained by appealing to the specific notion of deictic tense, contrary to Chung K.-S.’s (2007) argument. Criticizing the same points, Lee J. (2010) tries to account for the semantics of -te- in terms of a rather standard semantics of epistemic modality. Let us begin with the following simple sentence containing -te-:

! 174

(61)

Pi-ka

o-Ø-te-la.

Rain-Nom

fall-Pres-te-Decl

‘[The speaker made a sensory observation that] it was raining.’ (Lee J. 2010: 7, ex. 12a)

Following Kratzer (e.g., 1977, 1991), Lee J. (2010) assumes two conversational backgrounds: the modal base and the ordering source.

(62)

Modal base: gives a set of accessible worlds Ordering source: imposes a particular ordering among the accessible worlds determined by a modal base (Lee J. 2010: 6)

Both the modal base and the ordering source can be defined in terms of sets of propositions, as in Kratzer (1991). Then, let us further assume that, in (61), the modal base S0(w,t) and the ordering source ST/DX(w,t) are as follows:

(63)

Two conversational backgrounds for (61) a. modal base S0(w,t) = {Water is falling to the ground at t, there’s the sound of water dripping outside at t,…} b. ordering source ST/DX(w,t) = {It’s the rainy season at t, the person who lives upstairs is on vacation at t, the water pipe of the speaker’s apartment was repaired a short time prior to t.} (Lee J. 2010: 8, ex. 22)

! 175

Following Portner (2009), based on the modal base and the ordering source, Lee J. (2010) defines the following function, which provides the possible worlds that the modal -te- actually quantifies over:

(64)

BEST(S0, ST/DX, w, t) is a function from world-time pairs to sets of worlds which are ranked according to ST/DX(w,t) among the worlds determined by S0(w,t) Namely: the worlds in BEST(S0, ST/DX, w, t) are those that are compatible with what the speaker knows based on his/her sensory observation in w at t, and are ranked according to his/her expectations about what the world w was/is/would be like prior to/at/after t. (Lee J. 2010: 9, ex. 26; emphasis original)80

Based on these notions, Lee J. (2010) defines the lexical entry for -te- as follows:

(65)

-te-: %p.%w.%t.&w'[(w''BEST(S0, ST/DX, w, t) " t < now) ' p(w')(t)] (Lee J. 2010: 9, ex. 27)

That is, a proposition with -te- is true if and only if, for every world accessible to the speaker and for the time interval prior to the time of the utterance, the prejacent is true in each world and in that time period. Lee J.’s (2010) proposal can account for the temporal semantics of -te- as well as its similarities to epistemic modality (because it IS the epistemic modal, according to her lexical entry). However, I would like to make two points. First, it is unclear in the lexical entry in (70) !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 80

As Lee J. (2010) agrees, the function BEST in (65) is introduced for simplicity: just like Kratzer (1977, 1991), in her analysis the epistemic modal -te- is evaluated in terms of the modal base and the ordering source.

! 176

how we can derive the apparent inferential evidentiality when it is used with the past tense -essor the modal morpheme -keyss-. Lee J. might defend her position by changing the two conversational backgrounds accordingly, but in this case, we do not know how to derive this kind of meaning compositionally. Second, it is unclear in her analysis how we account for the evidential perspective shift of -te- in questions, as illustrated in (50) and (51). To respond to all questions raised above, tentatively I propose the following lexical entry for -te-:

(66)

The lexical entry for -te- (final version) For any utterance context c*, any utterance time t* and any world w, [[-te-]] = %p.%c: !t < t* which is salient in c and at t, sc perceived direct evidence compatible with p. p81

According to (66), -te- introduces the evidential presupposition like -ta-, and it also introduces the presupposition that there is a salient time point t before the utterance time t*, and s* has direct evidence compatible with p. Since in this lexical entry -te- is treated as a function from propositions to characters, just like -ta-, the evidential perspective shift is accounted for in the same way as the reportative evidential -ta-.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 81

Even though in Chapter 3 we saw that there is no convincing evidence that the assertive component of -te- is that of epistemic modals, the epistemic modal meaning can be easily incorporated to my lexical entry for -te- in the following way: (66')

The lexical entry for -te- (modalized version) For any utterance context c*, any utterance time t*, and any world w, [[-te-]]c* = %pst.%c: Bc = {p: sc believes that p is true only based on direct evidence he/she got at a salient time point t in c, which is before the utterance time t*}. %w. &w'[w''Bc(w)*p(w')] (c is a variable over contexts, that is, the triple )

Since, as we saw in the end of Chapter 3, -te- seems to express the epistemic necessity the speaker has for the prejacent, here I treat -te- as universal quantifiers over accessible possible worlds.

! 177

Then let us see how this lexical entry for -te- further accounts for the puzzles about -tepresented above. First consider the case where -te- by default, which is repeated below for convenience from (53a):

(67)

John-i

ecey

Mary-lul

manna-te-la.

John-Nom

yesterday

Mary-Acc

meet-te-Decl

‘John met Mary yesterday.’ Implication: the speaker has direct evidence (e.g., the speaker saw) that John met Mary.

Following anaphoric theory of tense (Partee 1984, Higginbotham 2010, among others), in the prejacent, when there is no overt tense marker, then an anaphoric tense, tpro, is introduced instead. Then the logical form of (67) can be represented as (68) (I omit the temporal adverbial for convenience):

(68)

[-te-

[tpro

[%w. John meets Mary]]]

Since there is no overt tense marker preceding tpro, it should get its value from the context c, which is introduced by -te-: the presupposition of -te- indicates that there is a salient time point t before the utterance time t* when sc came to get the direct evidence relative to p. Given this, tpro gets its value from t in the presupposition of -te-, and the sentence is interpreted as past. Therefore, since both the evidential presupposition and the prejacent are anchored to the same temporal point, that is t in the evidential presupposition, we have the simple direct evidentiality reading from -te-. Now let us turn to the case where -te- occurs with the past tense -ess-. Such example is given in (69), repeated from (54b): ! 178

(69)

(Scenario: yesterday I found that John had Mary’s book, which I had seen before was in Mary’s home) ?John-i

kucey

Mary-lul

manna-ess-te-la.

John-Nom

the-day-before-yesterday

Mary-Acc

meet-Past-te-Decl

‘John met Mary the day before yesterday.’ Implication: based on the speaker’s inference, John met Mary the day before yesterday.

Assuming that tpro is replaced by the past tense marker, the logical form of (69) is (70):

(70)

[-te-

[Past

[%w. John meets Mary]]]

First of all, note that the lexical entry for -te- in (66) only indicates that the evidence the speaker in c directly perceives is compatible with the prejacent. This lexical entry allows a speaker to adopt -te- even when he/she does not have any direct evidence relative to the prejacent: as far as his/her direct evidence is compatible with the prejacent, he/she can use -te- to indicate his/her evidence relative to the prejacent. Furthermore, note that the time when the speaker got his/her evidence that John met Mary two days ago is still before the time of utterance, which is also compatible with the presupposition of -te-, that is, there is a time point t before the utterance time t*, when the speaker got his/her direct evidence relative to the prejacent. Then the question is, how to derive the apparent inferential evidentiality of -te- when it is used with the past tense marker -ess-? One plausible account is to contribute this reading to some kind of implicature triggered by -te-. In (74), suppose that the speaker directly perceived the event described by the prejacent, that is, John met Mary. Then the speaker would NOT use -te- to indicate this direct perception: since -te- just mentions that the direct evidence the speaker has is ! 179

only compatible with the prejacent, it is much shorter and stronger NOT to use -te- to indicate his/her direct evidence. Thus, when using -te- with the past tense marker, by this implicature, the speaker did not directly perceive the event described by the prejacent, but only has direct evidence which is compatible with the prejacent, as expected. Finally, consider the case where the future tense/modal marker -keyss- appears with -te-, the example of which is (71) (repeated from 54c):

(71)

(Scenario: John has some problem with his homework and needs some help as soon as possible. I think Mary is the only one who can help him, but I saw Mary’s schedule where she is out of town today and will be back tomorrow.) John-i

nayil-i-lato

Mary-lul

manna-keyss-te-la.

John-Nom

tomorrow-L-at least/even

Mary-Acc

meet-may-te-Decl

‘John will meet Mary tomorrow at the latest.’ Implication: based on the speaker’s inference, John will meet Mary tomorrow.

The logical form of (71) is (72):

(72)

[-te-

[Fut

[%w. John meets Mary]]]

The future tense marker in (72) indicates that the event described by the prejacent does not occur yet. Thus, since the speaker cannot directly perceive the event which does not occur yet, his/her direct evidence is only compatible with the event which might occur in the future. Hence, indirect/inferential evidentiality is triggered by -te-. Concluding this section, I argued that, just like -ta-, -te- should also be treated as a function from propositions to characters, to account for the evidential perspective shift in ! 180

questions. In addition to that, I also showed that the apparent variability of the evidential presupposition introduced by -te- can be accounted for by specifying that -te- only indicates that the evidence the speaker directly perceives is compatible with the prejacent. More detailed analysis might be required, of course, but I leave this for future research.

4.8. Conclusions

In this chapter, I showed that the evidential perspective in Korean questions is obligatorily shifted from the speaker to the addressee. I also proposed that this obligatory shift can be best accounted for by assuming that Korean evidentials are partial functions from propositions to characters, introducing expressive presuppositions. By doing so, I showed that we do not need to fundamentally revise Hamblin’s (1973) rather standard semantics of questions (unlike Garrett 2001) or Kaplan’s (1989) theory of indexicals in order to account for the evidential perspective shift. What is required is the specific lexical semantics of evidentials. I also discussed why previous approaches do not account for the Korean facts, and why the Korean evidential is different from other perspective-involving operators, such as expressives and sentential adverbs. Finally, I proposed that the direct evidential -te- should also be analyzed as a function from propositions to characters, just like -ta2-.

4.9. Further discussion: deriving a set of characters in Karttunen (1977)

In this section I adopt Hamblin’s (1973) semantics of questions to account for the evidential perspective shift. As mentioned in Chapter 1, in the last section of this chapter, I discuss how to derive the evidential perspective shift in questions in Karttunen’s (1977) ! 181

semantics, which may further tell us what the differences between two theories of interrogatives are, especially in their theoretical implications as well as in their empirical predictions. To achieve this goal, first let us consider a simpler case, that is, how to derive the denotation of a yes-no question without any evidential marker in Karttunen (1977). An example of such a case is given in (73):

(73)

John-i

na-lul

po-ass-ni?

John-Nom

I-Acc

see-Past-Q

‘Did John see me?’

One major difference between Hamblin (1973) and Karttunen (1977) is that, whereas Hamblin (1973) assumes a bunch of rules which allow us to combine different types of arguments and functions in a set-tolerant way, as we saw above, Karttunen (1977) assumes a question morpheme, which takes a proposition (that is proto-question) and returns a set thereof.82 The lexical entry for this question morpheme is given in (74):

(74)

[[Q]] = %pst. {p}

Furthermore, to derive an extension of a yes-no question, Karttunen (1977), following Hamblin (1973), assumes that every yes-no question includes (either overt or covert) whether, which takes a singleton set of proposition derived by a question morpheme and returns a set of two propositions, corresponding to the affirmative answer and the negative answer. Its lexical entry is (75): !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 82

Karttunen (1977) also assumes special quantifiers and rules for wh-words, but for simplicity I do not discuss these aspects in this section.

! 182

(75)

[[whether]] = %Q. {p: p'Q or ~p'Q}

Given this, the LF of (73) can be represented as LF in Figure 4.6., where the proposition that John saw me - that is the proto-question - combines with the question morpheme resulting in a set containing that proposition, and that set combines with whether, resulting in a set of two propositions, corresponding to the affirmative answer and the negative answer:

Figure 4.6. LF of (73) John-i na-lul po-ass-ni? in Karttunen (1977) { %w. John saw me in w, %w. John did not see me in w} q p %Q. {p: p'Q or ~p'Q} { %w. John saw me in w} q p %p.{p} 6 %w. John saw me in w Given this, let us return to an example of a question with an evidential marker, that is (76) (repeated from above):

(76)

John-i

na-lul

po-ass-ta-ni?

John-Nom

I-Acc

see-Past-ta-Q

‘Did John see me?’ Implication: The addressee is expected to have reportative evidence about whether John saw the speaker.

! 183

As I illustrated above, to derive the evidential perspective shift, the extension of (76) should be a set of two characters, corresponding to the affirmative and the negative answer. This set is given below as (77):

(77)

{%c: sc has reportative evidence that John saw s*. %w. John saw s* in w, %c: sc has reportative evidence that John did not see s*. %w. John did not see s* in w}

Now let us consider how previous lexical entries can be changed to derive (77), while following each step in the computation. First, we have the proposition that John saw me, the denotation of which is (78):

(78)

%w. John saw me in w

Then we need to decide whether the question morpheme or -ta- takes wide scope. Consider the alternative where -ta- takes wide scope over the question morpheme:

(79)

[whether

[-ta-

[Q

[%w. John saw me in w]]]]

In this case we do not need to modify the lexical entry for the question morpheme, but note that the LF like (79) is not interpretable, due to the following reason. We have been assuming that -tais a function from propositions to characters. However, if the question morpheme applies to (78) before -ta-, the result of his application is a set of propositions. Since we cannot use the settolerant functional application rules any more in Karttunen’s (1977) semantics of questions, there

! 184

is no way for -ta- to apply to the element of this set, no matter what the question morpheme looks like. Therefore -ta- should combine first, resulting in a character with a definedness condition:

(80)

[[-ta-]](%w. John saw me in w) = %c: sc has reportative evidence that John saw me. %w. John saw me in w

Since this is a character rather than a proposition, our previous lexical entry for the question morpheme given in (74), which takes a proposition as its argument, cannot directly combine with this character. One way to solve this problem is to assume a different lexical entry for the question morpheme, which is given in (81), where it takes a character as its argument and returns a singleton set containing it:

(81)

[[Q]] = %k. {k}

Similarly, in the previous lexical entry, whether takes a set of propositions rather than a set of characters, and therefore it cannot combine with the result of applying (81) with (80), which is a set of characters. To allow whether to take a set of characters, the lexical entry for whether should be changed as follows:

(82)

[[whether]] = %Q<,t>. {k: k'Q or ~k'Q}

Finally, the lexical entry for the negation operator ~ in (82), which is commonly regarded as a propositional operator (that is, it takes a proposition as its argument and returns a proposition),

! 185

should also be modified accordingly, to be applied to a character. This modification can be done in the following way:

(83)

[[~]] = %k. %c. %w. k(c)(w) = 0

With these modified lexical entries, the logical form of (81) can be represented as in Figure 4.7:

Figure 4.7. LF of (76) John-i na-lul po-ass-ta-ni? in Karttunen (1977) {%c: sc has rep. evi. that John saw me in w. %w. John saw me in w, %c: sc has rep. evi. that John did not see me in w. %w. John did not see me in w } q p %Q<,t>. {k: k'Q or ~k'Q} {%c: sc has rep. evi. that J. saw me. %w. J. saw me in w} q p %k.{k} 6 c %c: s has rep. evi. that J. saw me. %w. J. saw me in w Now we can see that, in Karttunen’s (1977) semantics of questions, there are more restrictions on lexical entries for the question morpheme as well as other components, including whether (and possibly other wh-words) and negation. In contrast, as we saw above, in Hamblin’s (1973) semantics of questions there is no need to modify any lexical entry or any rule. In addition to that, there is other potential problem in the logical form in Figure 4.7. Note that, in this logical form, each function in whether takes a character with a definedness condition, deriving the affirmative answer and the negative answer. However, taking a closer look at the negative answer, we can see that with negation the definedness condition of the negative answer should also be changed accordingly (that is, from the condition that sc has reportative evidence that John saw me to the condition that sc has reportative evidence that John did not see me), which is rather unexpected in the presupposition projection. In Hamblin (1973) we do not need to care for this problem, since in this case -ta- takes wide scope over whether. ! 186

Summarizing discussions above, there are several reasons to adopt Hamblin’s (1973) semantics of questions rather than that of Karttunen’s (1977): its theoretical simplicity as well as its consistency with the previous theory of presupposition projection. There might be other ways to solve this problem in any modified theory of Karttunen’s semantics of questions, but at the moment, I do not see any possible solution, and therefore I will continuously adopt Hamblin (1973).

! 187

Chapter 5 Remaining Issues and Further Implications

In the previous chapters, I claimed that there are two types of evidential markers in Korean: the direct perceptive evidential -te- and the indirect/reportative evidential -ta- (Chapter 2). I also demonstrated that these markers show similarities to epistemic modals and that their evidential component is a presupposition (Chapter 3). Finally, I observed that one of the core features of evidentiality is indexicality, that is, the presuppositions that evidentials trigger vary depending on who the speaker is in the utterance context. However, Korean evidential markers and other indexicals are different in one crucial respect: presuppositions introduced by evidentials in questions are ultimately anchored to the addressee, rather than to the speaker, whereas no such shift is observed with other Korean indexicals. In Chapter 4, I suggested that as far as Korean evidentials are concerned, this shift should be accounted for in a semantically compositional way, because of its obligatory nature. Specifically, I proposed that evidential markers in Korean denote functions from propositions (of type ) to characters (of type ), and I also showed that the evidential perspective shift follows from this assumption alone, without any significant modification to the standard semantics of questions or other indexicals. Concluding this thesis, I address several remaining questions and further issues that my proposal raises. Whereas for some of these questions, there are obvious and compelling answers, for others, my conclusions have to be more tentative. I am convinced, however, that the discussion, though in part speculative, will prove helpful at least in providing a general direction for future research on the topic. The questions I deal with in this concluding chapter include the role of Korean sentential endings and their interactions with evidential markers (Section 5.1), the ! 188

‘affectionate’ or ‘intimacy’ reading of Korean reportative evidentials (Section 5.2), evidential uses of speaker-oriented adverbials across languages (Section 5.3), cross-linguistic variation in evidentials with respect to the evidential perspective shift (Section 5.4), and reportative evidentiality in the Korean noun complements (Section 5.5). Section 5.6 provides a conclusive summary of this dissertation.

5.1. Sentential endings and evidential markers

In this section, I consider the relationship between Korean sentential endings and evidential markers. Specifically, I discuss some further implications that Kaplan’s (1989) twostep semantics carries when applied to sentential endings in Korean.

5.1.1. Review of Kaplan (1989)

Recall that according to Kaplan’s (1989) semantics of indexicals, a grammar generates sentences whose denotations are characters, i.e., functions from contexts to intensions, which in turn are functions from worlds of evaluation to extensions. Under this schema, every sentence is assumed to be interpreted via two steps. First, a character applies to a context, resulting in an intension relative to that context. Then, the intension applies to a world of evaluation, resulting in an extension of the sentence relative to that context and world of evaluation. In terms of typedriven semantics, characters and intensions can be formalized as in (1), which I repeat from the previous chapter for convenience.

! 189

(1)

(revised from von Fintel 2005) a. The character of (:

[[(]]) = %c.%w.[[(]]c,w

b. The intension of ( at context c: [[(]]c¢ = %w.[[(]]c,w

Under this approach, every sentence is interpreted relative to two parameters: the context parameter and the world of evaluation parameter. The utterance context c* is defined as a triple of the world of the utterance w*, the time of the utterance t*, and the speaker of the utterance s*.

(2)

Utterance context c* =

Now, let us review how this system accounts for the context-dependency of indexicals in more formal terms. Consider the following sentence:

(3)

The president of GM must have seen me.

Under Kaplan’s (1989) two-step semantics, the denotation of this sentence is relativized to two parameters: the context variable c and the world-of-evaluation variable w.

(4)

[[The president of GM must have seen me]]) = %c.%w.[[The president of GM must have seen me]]c,w

Therefore, when a speaker utters (3), he/she applies (4) to his/her utterance context c*, resulting in the intension in (5).

! 190

(5)

[[The president of GM must have seen me]]¢ (%c.%w.[[The president of GM must have seen me]]c)(c*) = %w.[[The president of GM must have seen me]]c*,w

The intension of (3), which is given in (5), then applies to the world of evaluation. In this case, the epistemic modal must quantifies over the epistemic alternatives accessible from w in c*. Therefore, (3) is true if and only if, for every alternative world epistemically accessible from w, the complement proposition that the president of GM in THAT WORLD saw the speaker s* is true, that is:83

(6)

For every c* and w, [[The president of GM must have seen me]]c*, w = &w' accessible from s* in w* in c*: [the president of GM saw s* in w' in c*]

As noted above, the difference in interpretation of indexicals versus definite descriptions follows straightforwardly. While indexicals receive their denotations directly from the utterance context, the interpretation of definite descriptions depends on the world of evaluation, which may vary in its reference when the world variable is quantified over. In the discussion below, however, it is important to note that to evaluate a sentence in a two-step semantics, the character is regarded as containing a context variable c which is bound by a lambda operator. In the same way, we need to introduce a world variable parameter bound by a lambda operator in order to understand intensional phenomena in natural language.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 83

This is the so-called de dicto reading of the definite. As before, I ignore the de re reading, where the modal operator takes narrow scope with respect to the definite.

! 191

5.1.2. Back to evidentials in declaratives

Recall the previous examples, where the reportative evidential marker -ta- is used in a declarative and in a question.

(7)

a. John-i John-Nom

na-lul po-ass-ta-nta. I-Acc

see-Past-ta-Decl

‘John saw me’ Presupposition: the speaker has reportative evidence that John saw the speaker. b. John-i John-Nom

na-lul po-ass-ta-ni? I-acc

see-Past-ta-Q

‘Did John see me?’ Presupposition: the speaker expects the addressee to have reportative evidence about whether John saw the speaker.

The central ingredient of the account for the evidential perspective shift that I proposed in the previous chapter is the assumption that Korean evidentials are functions from propositions to characters. Therefore, the intension of (7a) and (7b) is computed as (8a) and (8b), respectively.

(8) a. [[(7a)]]c* = [[-ta-]](%w. John saw s* in w) = [%p. %c: sc has reportative evidence relative to p.p] (%w. John saw s* in w) = %c: sc has rep. evi. relative to the proposition that J. saw s*. %w. J. saw s* in w

! 192

b. [[(7b)]]c* = [[-ta-]]([[whether John saw s*]]) = {%c: sc has rep. evi. relative to the proposition that J. saw s*. %w. J. saw s* in w, %c: sc has rep. evi. relative to the proposition that J. didn’t see s*. %w. J. didn’t see s* in w}

As the reader might have already noticed, the intensions of these two sentences are no longer functions from worlds to extensions, but a character in one case and a set of characters in the other. Given this, a serious problem emerges in the derivation of (7a), that is, (8a). In (8a), for the pronoun to be interpreted, the intension of the sentences is derived by relativizing the interpretation function to a specific context, namely, the utterance context c*. Note, however, that the result of the derivation is not an intension as in Kaplan’s theory, but instead still has the semantic type of a Kaplanian character. In order to obtain an intension from it, another application to the utterance context c* is required, which is not predicted. To put this problem differently, consider the character of (7a) and (7b), which is shown below as (9a) and (9b), respectively. In (9a), even though the character is applied to the utterance context c*, we still get a character rather than an intension, which is problematic.

(9) a. [[(7a)]]) = %c'. [[(7a)]]c' = %c'. %c: sc has rep. evi. relative to the proposition that John saw sc'. %w. John saw sc' in w b. [[(7b)]]) = %c'. [[(7b)]]c' = %c'.{%c: sc has rep. evi. relative to the proposition that J. saw sc'. %w. J. saw sc' in w, %c: sc has rep. evi. relative to the proposition that J. didn’t see sc'. %w. J. didn’t see sc' in w}

! 193

Compare (7a) with the corresponding question, which is (7b). In questions, my proposal accounts for the evidential perspective shift without any problem. If the speaker s* in the context c* utters (7b) in the world-of-evaluation w*, then (9b) is applied to the context c*, resulting in (10).

(10)

[[(7b)]]c* = (%c'. [[(7b)]]c')(c*) = [%c'.{%c: sc has rep. evi. relative to the proposition that J. saw sc'. %w. J. saw sc' in w,

%c: sc has rep. evi. relative to the proposition that J. didn’t see sc'. %w. J. didn’t see sc' in w}](c*) = {%c: sc has rep. evi. relative to the proposition that John saw s*. %w. John saw s* in w, %c: sc has rep. evi. relative to the proposition that J. didn’t see s*. %w. J. didn’t see s* in w}

Here, the speaker uttering the question, that is, s* does not directly utter any of the answers in (10b), and therefore, the evidential presupposition is expected to be anchored to the addressee, whereas the pronoun is expected to be anchored to s*. Let us now return to (7a). If s* utters (7a), then (9a) is applied to c*, resulting in (11).

(11)

[[(7a)]]c* = (%c'. [[(7a)]]c')(c*) = (%c'. %c: sc has rep. evi. relative to the proposition that J. saw sc'. %w. J. saw sc' in w)(c*) = %c: sc has rep. evi. relative to the proposition that J. saw s*. %w. J. saw s* in w

Here is a more theory-internal problem: What we get after the application of (7a) to c* is still a character, where the evidential presupposition is anchored to the context variable c, which is still bound. However, recall from the previous chapter that what we ultimately need is for the evidential presupposition to be anchored to s*. But, for this to be the case, the character in (11) should apply to c* again, an operation that is not envisaged in the Kaplanian system.

! 194

We saw above that there are obvious problems with the proposal made in the previous chapter, and therefore, unless something else in the sentence provides an adequate context argument to the character in declaratives with evidential markers, these problems will remain unresolved.

5.1.3. Semantics of declarative endings

What has not been discussed so far, and which can solve the problems mentioned above, is the semantic role of Korean declarative endings. As the reader may recall from previous chapters, every Korean sentence contains a sentential ending that indicates whether the sentence is a declarative, an interrogative, an imperative, an exhortation, a promise, or an exclamation. These markers are obligatory in Korean, as (12) illustrates.

(12)

*John-i

Mary-lul

ttayli-ess.

John-Nom

Mary-Acc

beat-Past

Intended: John beat Mary.

For the moment, let us focus on the declarative endings. The main role of declarative endings is, of course, to indicate that a sentence is declarative. However, they also indicate in which context the given sentence can be uttered felicitously. This is demonstrated in (13), where the declarative endings are shown to vary depending on the register and social status of the conversation participants.

! 195

(13) a. John-i John-Nom

Mary-lul

ttayli-ess-ta.

Mary-Acc

beat-Past-Decl

‘John beat Mary’ Implication: the register is plain, and the speaker’s social status is higher than the addressee’s. b. John-i John-Nom

Mary-lul

ttayli-ess-e.

Mary-Acc

beat-Past-Decl

‘John beat Mary’ Implication: the register is informal, and the speaker and addressee have the same social status. c. John-i John-Nom

Mary-lul

ttayli-ess-supnita.

Mary-Acc

beat-Past-Decl

‘John beat Mary’ Implication: the register is formal, and the speaker’s social status is lower than the addressee’s.

As we saw in previous chapters, all of the examples in (13) are truth-conditionally equivalent and differ only in their felicity conditions relative to social aspects of the utterance context. The problem presented in the previous subsection is resolved if the semantic role of declarative endings also includes the introduction of an adequate context variable for the denotation of their argument (which was presented above) to apply to. In fact, this assumption might not be entirely unreasonable if the felicity conditions these endings introduce can themselves be seen as expressive presuppositions (like evidentials), specifying the register that the sentence should use and the relative social status of the speaker and addressee.

! 196

More concretely, I suggest that the intensions of Korean declarative endings are functions from characters to intensions which, among other requirements, have the function of applying the character in their argument position to the context of utterance they are interpreted relative to. Specifically, for any utterance context c*:

(14)

[[Decl]]c* = %k. reg-status(c*) & c* ' Dom(k). k(c*) (where k is a variable over characters, reg-status(c*) is the abbreviation of the definedness condition regarding the register and the social status of the participants that c* should satisfy)84

That is, for any utterance context c*, a Korean declarative ending takes a character and returns an intension, imposing a definedness condition (i.e., presupposition) about the register and social status of the participants on the utterance context c*. When a declarative ending applies to a character k, this character k also applies to the utterance context c*, with the definedness condition of the character k projected with the definedness condition on c*. Now, let us see how (14) accounts for the evidential marker in a declarative sentence. Consider (7a), which is repeated below as (15).

(15)

John-i

na-lul

po-ass-ta-nta.

John-Nom

I-Acc

see-Past-ta-Decl

‘John saw me’ Presupposition: The speaker has reportative evidence that John saw the speaker.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 84

It is beyond the scope of this thesis to formalize the presupposition reg-status(c*). Interested readers may refer to Pak et al. (2005, 2007), where the semantic implications of Korean imperative, exhortative, and promissive endings are analyzed in terms of presuppositions.

! 197

Since the declarative ending -nta is used when the register is plain and the social status of the speaker is higher than that of the addressee, as shown in (13a), let us tentatively define the lexical entry for -nta as in (16).

(16)

For any context of utterance c*, [[nta]]c* = %k: plain(c*) & s*>a* & c* ' Dom(k). k(c*) (where plain(c*) means that the register in c* is plain, and s*>a* means that the social status of the speaker s* in c* is higher than that of the addressee a* in c*)

With (16), the LF of (15) is as in Figure 5.1.85

Figure 5.1. LF of (15) John-i na-lul po-ass-ta-nta with declarative ending -nta (iii) 3(ii) -nta 3 -ta3 John 3 saw me (i) Then for any utterance context c*, the LF in Figure 5.1. is computed as follows:

(17)

(i) [[me]]c* = s*. (ii) %c: sc has reportative evidence that John saw s*. %w. John saw s* in w (iii) [[-nta]]c*([[-ta-]](%w. John saw s* in w)) = [[-nta]]c*(%c: sc has reportative evidence that John saw s*. %w. John saw s* in w)

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 85

For convenience, I draw the LF as head-initial, although Korean is a head-final language.

! 198

= [%k: plain(c*) & s*>a* & c* ' Dom(k). k(c*)] (%c: sc has rep. evi. that J. saw s*. %w. J. saw s* in w)

(17iii) is defined if and only if the utterance context c* is plain and the social status of the speaker is higher than that of the addressee (from the definedness context of -nta) AND if c* is in the domain of the argument character, that is, if the following also holds:

(18)

c* ' Dom(%c: sc has reportative evidence that John saw s*. %w. John saw s* in w).

When defined, then:

(17)

(iii) = %w. John saw s* in w.

Note that node (ii) still denotes the partial character, where the definedness condition is anchored to the bound variable c. When this character is combined with the declarative ending -nta, the definedness condition introduced by the non-final -ta- (i.e., the reportative evidential implication) is projected, and the evidential presupposition, as well as the presupposition for the register and the relative social status of the speaker and addressee added by -nta, then all become anchored to the utterance context c*. Although plausible, the solution I suggest in this section raises one additional issue. As the reader might have already noticed, the proposed semantics for declarative endings is based on the semantics of their complements, which is derived only if these sentences contain evidentials. Given this, the proposal appears inadequate for the derivation of the meaning of declarative sentences that do not contain any overt evidential marker. One plausible way of avoiding this problem is to assume that every Korean sentence contains an overt or covert evidential marker. ! 199

Thus, when there is no overt evidential marker, there is a covert evidential marker +, with a meaning similar to what Faller dubbed a ‘best possible grounds’ evidential, as sketched in (19).

(19)

[[+]]c* = %pst.%c: sc bases his/her assertion on the best possible evidence available to him/her. p (c is a variable over contexts, that is, a triple of ))

That is, the speaker sc uses the best possible evidence relative to p available to him/her, no matter what the type of the evidence might be. At first, this proposal seems too far-fetched, but I think it is still plausible because in many languages with evidential morphemes, if evidentials are not used in a sentence, then the sentence carries the implication that the speaker asserts the propositional content of the sentence based on his/her best possible grounds (as Faller illustrates for Cuzco Quechua).

5.1.4. Semantics of question endings

Just as declaratives require a declarative ending, questions require a question ending, which marks the sentence as question and introduces implications about the register and the participants’ relative social status, as illustrated in (20).

(20) a. John-i John-Nom

Mary-lul

ttayli-ess-ni?

Mary-Acc

beat-Past-Q

‘Did John beat Mary?’

! 200

Implication: the register is plain, and the speaker’s social status is higher than the addressee’s. b. John-i John-Nom

Mary-lul

ttayli-ess-e?

Mary-Acc

beat-Past-Q

‘Did John beat Mary?’ Implication: the register is informal, and the speaker and addressee have the same social status. c. John-i John-Nom

Mary-lul Mary-Acc

ttayli-ess-supnikka? beat-Past-Q

‘Did John beat Mary?’ Implication: the register is formal, and the speaker’s social status is lower than the addressee’s.

To account for the implication introduced by the question endings, and to make questions denote characters that ultimately apply to the utterance context c*, I suggest that Korean question endings also introduce the presupposition regarding the register and the relative social status of the participants.

(21)

For any utterance context c*, [[QUEST]]c* = %k<,t>: reg-status(c*) & c* ' Dom(k). k

The only difference between question endings and declarative endings is that, whereas the latter combine with a character and return an intension with the definedness condition regarding social aspects of the context, the former combine with a set of characters and return the intension of that question, that is, a set of characters introducing the definedness condition regarding the social ! 201

aspects of the context, as above. Therefore, the intension of ni is as in (22), where the definedness condition is the same as that of -nta.

(22)

For any utterance context c*, [[ni]]c* = %k<,t>: plain(c*) & s*>a* & c* ' Dom(k). k

Now, let us see how this works in a case like (23).

(23)

John-i

na-lul po-ass-ta-ni?

John-Nom

I-acc

see-Past-ta-Q

‘Did John see me?’ Presupposition: the speaker expects the addressee to have reportative evidence about whether John saw the speaker.

Including the question ending, the logical form of (23) becomes the LF in Figure 5.2.

Figure 5.2. LF of (23) John-i na-lul po-ass-ta-ni? with question ending -ni (v) 3 (iv) ni 3(iii) -ta3 (ii) whether 3 Johne 3 saw me (i) Up to node (iv), the denotation of (23) is derived as follows, as we saw in the previous chapter.

! 202

(24)

For any context c* () (i) [[me]]c* = s* (ii) [[John saw me]]c* = %w. John saw s* in w (iii) two functions in [[whether]]c* are applied to the proposition that J. saw s* via PFA: {%w. John saw s* in w, %w. John did not see s* in w} (iv) [[-ta-]]c*is applied to each proposition in (iii) via PFA: {%c: sc has reportative evidence that John saw s*. %w. John saw s* in w, %c.: sc has reportative evidence that John did not see s*. %w. John did not see s* in w}

When -ni applies to the set in node (iv), at node (v) we have the following denotation:

(25)

(v) = [[ni]]([[(iv)]]) = [%k<,t>: plain(c*) & s*>a* & c* ' Dom(k). k] ({%c: sc has reportative evidence that John saw s*. %w. John saw s* in w, %c.: sc has reportative evidence that John did not see s*. %w. John did not see s* in w})

For any context c*, (25) is defined if and only if the utterance context c* is a plain context and the speaker s* has higher social status than the addressee. If defined, then the intension of (25) in c* is as in (26), which is the same semantic object we saw in the last chapter.

(26)

{%c: sc has reportative evidence that John saw s*. %w. John saw s* in w, %c: sc has reportative evidence that John did not see s*. %w. John did not see s* in w}

! 203

The last remaining question is how to account for questions without any overt evidential marker. As with declaratives, I propose that every Korean question includes an evidential marker, either overt or covert, and that the covert default marker is BPG, which we saw above.

5.1.5. Interim summary

In this section, I discussed the semantics of declarative endings and question endings and claimed that declarative endings denote functions from characters to propositions that have, among other functions, that of applying characters to sets of characters whose only purpose is to mark the utterance as a question and introduce expressive presuppositions relative to the contextual register and the relative social status of the speaker and addressee.

5.2. The ‘intimacy’ reading of Korean reportative evidential -ta-

In some contexts, the declarative ending with non-final -ta- seems to carry implications different implications from those of the indirect evidential. One example is in the context of story-telling and fairy tales (Ji Young Shim, p.c.). Consider (27).

(27)

Yeysnalyeysnaley

etten wangkwuk-ey

wanca-wa

kongcwu-ka

Once upon a time

certain kingdom-Loc

prince-Conj

princess-Nom

sal-ko iss-ess-ta-nta. live-Prog-Past-ta-Decl ‘Once upon a time in a certain kingdom, there lived a prince and a princess.’

! 204

In (27), by using -tanta, the speaker seems to express his/her intimacy with the addressee and make the addressee comfortable. I still think, however, that this use of -tanta can be captured in terms of the reportative evidentiality of -ta- if the following two characteristics of reportative evidentials are clarified: First, evidentials do not specify the exact source of information, only the type of information (perception, hearsay, inference, etc.), and second, when using the reportative evidential -ta-, the speaker indicates that he/she does not have direct evidence, thus distancing himself/herself from the event described. Under this interpretation, the reportative evidential is felicitous in storytelling or fairy tale contexts, since the speaker is simply telling stories which he/she heard from someone else (the speaker does not need to care about the identity of the source). In this sense, the speaker can indirectly express that he/she is close in his/her experience to the addressee. The ‘intimacy’ flavor of -tanta can also be accounted for by appealing to the lesser degree of certainty expressed by the reportative evidential -ta-. Since it makes the speaker’s assertion ‘less strong’ or indicates that the speaker only makes a weak commitment to the truth of the story, the addressee may feel more comfortable when he/she hears a new story told by the speaker. Given this, the ‘intimacy’ flavor of -tanta is treated only as a pragmatic by-product of the core function of the reportative evidential. However, not all ‘intimacy’ readings of the reportative evidential -ta- can be accounted for by the pragmatic reasoning illustrated above. Consider the following example (Min-Joo Kim p.c.):

(28)

(A mother says to her daughter, Mary) Emma-nun

Mary-lul

salangha-n-ta-nta.

Mom-Top

Mary-Acc

love-Pres-ta-Decl

‘I love you’ (literally: Mommy loves Mary)

! 205

(28) is perfectly felicitous to Korean native speakers. Here, the mother (that is the speaker) is making a strong commitment to the fact that she loves her daughter, Mary, by using -tanta. It is ridiculous to think that the mother has only reportative evidence that she loves her daughter (which is what we would expect if -ta- in -tanta were always a reportative evidential). The mother has direct evidence that she loves Mary, because she herself is the one who loves Mary. One easy way to account for examples such as (28) is to treat -tanta as a declarative ending introducing a different implication. For example, Min-Joo Kim (p.c.) claims that the complex declarative ending -tanta should be treated differently from other declarative endings introducing the reportative evidential implication, such as -tay or -tapnita. However, we still need an account for -tanta with the reportative evidential implication, as we have seen throughout this thesis. Furthermore, to treat -tanta simply as a different ending does not explain why -tapnita also has an intimacy reading, as we will see later. The next possible analysis is to assume that there are two different -tantas, as do most Korean linguists. Here, let us consider Chae’s (2006) analysis. Chae (2006) claims that two types of -tanta should be distinguished: the ‘non-reducible’ -ta-nta and the ‘reducible’ -tanta. According to him, ‘reducible’ -tanta can be further analyzed as containing a covert quotation verb, corresponding to say (this analysis was described in Chapter 2). Chae (2006) also claims that nonreducible -ta-nta is formed by attaching the derivational suffix -nta, a homophone of the declarative ending -nta, to the declarative ending -ta. Roughly speaking, reducible -tanta corresponds to a -ta-nta that includes the reportative evidential -ta-, whereas non-reducible -tanta corresponds to a -tanta with the intimacy reading in (28) and (29a) (and perhaps the -tanta in 28). Under Chae’s analysis, the difference between reducible -tanta and non-reducible -tanta lies in the ambiguity of the -nta in -tanta (which is either a declarative ending or a derivational suffix). However, this analysis also has several problems. First, as shown in (29a) and (29b), since -tapnita also exhibits the intimacy reading, we have to assume ambiguity in the ending ! 206

pnita in (29b) as well as in the ending -nta, which is not so plausible. Furthermore, even if we assume the ambiguity of -pnita and -nta, this analysis cannot account for why -tay lacks the intimacy reading. Consider three endings, including the reportative evidential -ta-, that we saw in previous chapters.

(29) a. Nay-ka I-Nom

ne-lul

salangha-n-ta-nta.

you-Acc

love-Pres-ta-Decl

tangsin-ul

salangha-n-ta-pnita.

you-Acc

love-Pres-ta-Decl

ne-lul

salangha-n-ta-e. (=> tay)

you-Acc

love-Pres-ta-Decl

‘I love you’ b. Cey-ka I-Nom ‘I love you’ c. Nay-ka I-Nom ‘I love you’

The implications of (29a) and (29b) are ambiguous between two readings.86 Under the evidential reading, these two sentences carry the implication that the speaker heard from someone else that the speaker loves the addressee, and as we saw in Chapter 2, there is an additional implication that the speaker does not make much commitment to the truth of the prejacent (i.e., we have the first-person effect; if the relationship between the speaker and addressee is very close, as in (28), this reading is unavailable). With the intimacy, reading (29a) and (29b) express the speaker’s intimate and affectionate feelings toward the addressee. However, in (29c), we do not have the intimacy reading of -ta-; (29c) only carries the reportative evidential implication. Now, if we assume that -ta and -pnita are ambiguous between the declarative ending and the suffix, it is !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 86

In this section, I do not consider the ‘real’ contraction of the quotation, which I discussed in Chapter 2.

! 207

natural to assume that the ending -e in (29c) is also ambiguous between the declarative ending and the suffix and, therefore, the reportative evidential reading and the intimacy reading. However, as have seen, (29c) has only the first reading, which means that Chae’s (2006) hypothesis that the sentential ending -nta is ambiguous between the declarative ending and the derivational suffix is not particularly plausible. The other possibility is to assume the ambiguity of non-final -ta2- between the reportative evidential -ta2- and the other -ta- (whatever it might be: a declarative or other type of ending). This is not plausible, either. Consider the non-final -ta2- in questions, which is repeated below for convenience.

(30) a. John-i John-Nom

Bill-ul

ttayli-ess-ta-ni?

Bill-Acc

beat-Past-ta-Q

‘Did John beat Bill?’ Implication: the addressee heard from someone else whether John beat Bill. b. John-i John-Nom

Bill-ul

ttayli-ess-ta-pnikka?

Bill-Acc

beat-Past-ta-Q

‘Did John beat Bill?’ Implication: the addressee heard from someone else whether John beat Bill c. John-i John-Nom

Bill-ul

ttayli-ess-ta-e? (=> ttayli-ess-tay)

Bill-Acc

beat-Past-ta-Decl

‘Did John beat Bill?’ Implication: the addressee heard from someone else whether John beat Bill

All of these examples have only evidential readings. For example, (30a) does not carry the implication that the questioner is in a close relationship with the answerer. Rather, it carries only ! 208

the evidential presupposition that the speaker expects the addressee to have reportative evidence about to whether John beat Bill. If the intimacy reading were due to the ambiguity of -ta2-, there would be no reason for it not to appear in questions. Summarizing the discussions above, none of the accounts assuming any ambiguity in the lexicon seems so plausible. To account for the intimacy reading of -ta2- without assuming any ambiguity, let us assume two other possibilities. The first is an account based on the pragmatics of -ta2-, which can be described in the following way:87 The speaker would like to express his/her intimacy by using -ta-, thus expressing less certainty about the assertion, as in (27). However, using -ta- with the first-person pronoun produces the first-person effect that the speaker is not aware of the event described by the prejacent (see Chapter 2, as well as Aikhenvald 2004: Ch. 7), which is non-sense with a psych-predicate such as salangha- ‘love’. To avoid this kind of pragmatic conflict, the speaker denotes himself/herself as well as the addressee in the third-person, thus distancing himself/herself from the event described by the prejacent (e.g., loving her daughter in 30). By doing this, the speaker can express his/her intimacy to the addressee using ta- and avoid the first-person effect. The implausibility of the intimacy reading of -tay is explained by appealing to another kind of pragmatic reasoning, such as: ‘since the declarative ending -e in the -tay form already expresses that the speaker and the addressee are in a close relationship, the use of -ta- to express intimacy is redundant.’ This kind of account for the intimacy reading of -ta- seems empirically incorrect, however. This account makes the following two predictions: first, the intimacy reading of -tashould NOT be available when the speaker uses the first- or second-person pronoun; and second, the intimacy reading of -ta- is available only when the predicate is a psych-predicate. Neither

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 87

Maria-Luisa Zubizarreta (p.c.) makes a similar suggestion, although not exactly the same as what is illustrated in this paragraph.

! 209

prediction is borne out; (31) is a counterexample to the first prediction, and (32) to the second (and actually the first, as well).

(31)

Na-nun

ne-lul

salangha-n-ta-nta.

I-Top

you-Acc

love-Pres-ta-Decl

‘I love you’ Implication: the speaker expresses his/her intimacy to the addressee

(32)

Cey-ka tangsin-eykey ku kkochtapal-ul

senmwulha-ess-ta-pnita.88

I-Nom you-Nom

present-Past-ta-Decl

that bunch of flowers-Acc

‘I presented you that bunch of flowers (as a gift)’.

The other possibility is to analyze the ‘intimacy’ reading in terms of various pragmatic effects of modality, such as the ‘politeness’ implication of might and would. For example, “[might] is often used in questions as a more tentative, more polite, form for asking permission” (Palmer 2001: 74).

(33)

Might I come in at the moment on this, Chairman?

Would can also express politeness, as in the idiom would like...; thus, (34b) is generally regarded as a more polite expression than (34a).

(34) a. I want to take a rest. b. I would like to take a rest. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 88

Ce (or cey) and tangsin are the honorific forms of the first and second person pronouns, respectively.

! 210

Since modal operators generally introduce various types of pragmatic implications depending on different contexts, it seems plausible that the intimacy reading of the reportative evidential -taneeds to be analyzed in parallel with the politeness reading of some modal operators. There are also some problems with this, though. Most of all, we do not treat Korean evidentials as sort of epistemic modals (see Chapter 4), and even though we assume that Korean evidentials are a certain type of epistemic modal, the syntactic and semantic properties of epistemic modals differ from those of other so-called root modals. Furthermore, as far as I can see, epistemic modals do not introduce any additional pragmatic implications. Actually, both modal operators in (33) and (34b) are root modals and not epistemic modals; might in (33) is more likely to be a deontic modal, and would in (34b) seems to be volitional (Portner 2009: Ch. 4.4). Therefore, it seems less plausible for evidentials, which show similarities to epistemic modals, to introduce additional pragmatic implications. At the moment, I do not have a complete answer to this issue. I think the pragmatic account is the most plausible (even though it makes incorrect empirical predictions, such as 32 and 33), but I leave this issue for future research.

5.3. Evidentiality of sentential adverbials

Recall that there are various ways to express broad evidentiality in languages that do not have narrow evidentiality. For example, in Chapter 1, we saw that in English, broad evidentiality can be expressed through various means. The examples are repeated below.

! 211

(35) a. I saw that John beat Mary. b. Bill said, “John beat Mary.” c. Apparently John beat Mary. d. John must have beaten Mary.

One of the interesting cases is (35c), where sentential adverbials are used to express broad evidentiality. As pointed out in Chapter 1, sentential adverbials introducing broad evidentiality (for brevity hereafter I refer to such evidentials evidential adverbials), including apparently, reportedly, purportedly (and obviously?) are very similar to evidentials in languages that have narrow evidentiality. Since it is obviously beyond the scope of this chapter to provide a full analysis of those sentential adverbials, in this section, I only point out several issues of sentential adverbials involving the speaker’s perspective, and discuss similarities and differences between different types of sentential adverbials, focusing on the (apparent) distinction between adverbials which introduce broad evidentiality and adverbials which do not.89 Let us begin with evidential adverbials, especially purportedly. Before discussing the evidential implication of purportedly, however, note that it has two lexical meanings: one is similar to that of supposedly, and the other is similar to that of reportedly. This difference can be distinguished in the embedded context: when purportedly appears in the embedded clause, as in (36), it only has the meaning supposedly.

(36)

John knows whether John is purportedly a murderer.

Given this, consider the following example, where the evidential adverbial purportedly is used in a question: !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 89

For recent discussion and classification of the so-called speaker-oriented adverbs, see Ernst (2009).

! 212

(37)

Purportedly, who beat Mary?

I need to consult more native speakers, but my initial impression is that they do not show such ambiguity as we have with honestly. In my view, (37) only means that the addressee is expected to answer based on hearsay evidence (or based on what is rumored to the addressee, rather than to the speaker). In Korean, although it is difficult to find evidential adverbials corresponding to those in English, some derived adverbials also show the same perspective shift as in English. One example is given below.

(38) a. Somwun-ulo-nun Rumor-Ins-Top

John-i

Mary-lul

ttayli-ess-ta-e.

John-Nom

Mary-Acc

beat-Past-ta-Decl

nwu-ka

Mary-lul

ttayli-ess-ni?

who-Nom

Mary-Acc

beat-Past-Q?

‘According to a rumor, John beat Mary’ b. Somwun-ulo-nun Rumor-Ins-Top

‘According to a rumor, who beat Mary?’

Somwunulonun, the literal meaning of which is ‘according to rumor,’ introduces an implication similar to that of the hearsay evidential, namely, the speaker has heard from someone else (even though the exact information source is unclear) about what he/she asserts. In (39a), this implication is anchored to the speaker: according to the rumor the speaker has heard, John beat Mary. However, in (38b), it seems that it can be anchored only to the addressee, just like Korean evidentials. This shows that at least in questions, the behavior of Korean evidential adverbials parallels that of the reportative evidential. ! 213

This said, let us consider sentential adverbials. In Chapter 4, I observed that in Korean, the interpretation of sentential adverbials such as solcikhi ‘honestly’, when used in questions, can be anchored either to the speaker or to the addressee. The example is repeated below as (39).

(39)

Solcikhi,

John-i

ne-lul

manna-ess-ni?

Honestly

John-Nom

you-Acc

meet-Past-Q

‘Honestly, did John meet you?’ i) addressee-anchored: the speaker expects the addressee to answer honestly ii) speaker-anchored: the speaker is genuinely asking an information-seeking question (i.e., the speaker does not know anything about the answer)

Similarly, in English, as shown in Chapter 4, honestly is apparently ambiguous in questions: at first sight, sentential adverbials such as honestly seems to show the same ambiguity in questions. Consider the following example, which is the English counterpart of (39):

(40)

Honestly, did John meet Mary?

Unlike what is reported by Garrett (2001), among others, honestly in (40) seems to show the same ambiguity as in (39). On the one hand, honestly introduces the implication that the speaker expects the addressee to answer honestly (the addressee-anchored reading), but on the other hand, it may signal the speaker’s complete ignorance about what is asked/questioned (the speakeranchored reading). However, as mentioned in Chapter 4, this apparent ambiguity can be accounted for otherwise, especially if we note that in Korean as well as in English, solcikhi / honestly can be speaker-anchored only when the questioner is negatively biased to the answer, or, the questioner ! 214

would be really surprised at the positive answer (Andrew Simpson, Michael Shepherd, p.c.). This second implication can be explained if the apparent speaker-anchored reading of solcikhi / honestly is in fact another instance of the addressee-anchored reading. Suppose that the speaker is negatively biased. Then, by using solcikhi / honestly, the speaker asks the addressee to answer in an honest way, expressing the speaker’s skepticism about the positive answer. In this account, the apparent speaker-anchored reading can also be explained in terms of the speaker’s negative bias: the speaker does not expect that the answer is positive, and therefore he/she does not know how the answer can be positive. Given this, it seems that there is not much difference between evidential adverbials and other type of sentential adverbials. Other diagnostics identifying narrow evidentiality do not seem to distinguish evidential adverbials from other sentential adverbials. For example, these adverbials cannot appear in the embedded proposition. Consider English purportedly.

(41)

Mary knows whether John is purportedly a murderer.

As mentioned above, in this case purportedly cannot be used as an evidential adverbial, the meaning of which corresponds to that of reportedly. The Korean counterpart of (41), where somwunulonun apparently appears in the embedded clause, is simply unacceptable:

(42)

??Mary-nun Mary-Top

somwun-ulo-nun John-i

salinpem-incianinci

al-n-ta.

rumor-Ins-Top

murderer-whether

know-Pres-Decl

John-Nom

Similarly, solcikhi / honestly do not appear in the embedded clause. An English example (43) seems unacceptable. The Korean counterpart, (44), seems to be acceptable, but it is still

! 215

interpreted from the speaker’s perspective, and not from the matrix subject’s perspective, or from the embedded subject’s perspective.

(43)

??Tom thought that honestly John met Mary yesterday.

(44)

Tom-un solcikhi Mary-ka

Tom-ul

manna-ess-ta-ko

sayngkakha-n-ta.

Tom-Top honestly Mary-Nom Tom-Acc meet-Past-Decl-Comp think-Pres-Decl ‘Honestly, Tom thinks Mary met Tom’

Other diagnostics, including the question test and the negation test, seem to yield similar results, and do not distinguish one type of sentential adverbials from another (for brevity I do not repeat them in this section). I leave a detailed analysis of the difference between evidential adverbials and other adverbials for future research, but if the analysis provided in this thesis is on the right track, the perspective shift of evidential adverbials (and other sentential adverbials, probably) can also be accounted for in terms of the analysis of evidential markers (which take propositions as their arguments and return characters) provided with in this thesis. This kind of approach can be further extended to the difference between evidentials and other modal operators, which I discuss in the next section.

5.4. Evidentials and epistemic modals

In this section, I briefly discuss the difference between evidentials and epistemic modals, as well as their cross-linguistic variations. First, I discuss how evidentials and epistemic modals ! 216

are different in languages that show narrow evidentiality. Then, I make some speculations about cross-linguistic variations in narrow evidentiality and epistemic modality, focusing specifically on different aspects of the perspective shift in questions.

5.4.1. Epistemic modals and narrow evidentiality

Consider Kratzer’s (1991) semantics of modality. According to her, modals are quantifiers over possible worlds, and two different conversational backgrounds determine what possible worlds a given modal operator quantifies over. First, the modal base, which provides a set of possible worlds, varies depending on the interpretation of the given modal operator (deontic, epistemic, etc.). When the modal is deontic, the modal base provides a set of worlds that are compatible with the rules and obligations of the world of evaluation, and when the modal is epistemic, it provides a set of worlds that are compatible of the speaker’s knowledge in the world of evaluation (or compatible with the speaker’s evidence in the world of evaluation). Second, the ordering source further determines which world in the modal base is ‘closest’ to the world of evaluation. In terms of type-driven semantics, both the modal base and the ordering source can be formalized as functions from the world of evaluation to a set of propositions. Given this, the universal modal must and the existential modal may can be defined as follows (slightly modified from von Fintel and Heim 2007: 55):

(45) a. [[must]]w,g = %f>. %g>. %qst. &w''maxg(w)(.f(w)): q(w')=1. b. [[may]]w,g = %f>. %g>. %qst. !w''maxg(w)(.f(w)): q(w')=1.

! 217

(where f is the modal base, g is the ordering source, maxg(w) is the set of worlds closest to the world of evaluation in .f(w), and .f(w) is the set of worlds in which all of the propositions in f(w) are true)

Here, the modal must takes two sets of propositions, the modal base and the ordering source, as its conversational backgrounds, and applies to a proposition, resulting in the extension of q with respect to every world of evaluation selected by the ordering source in the modal base. Compare (45) with the lexical entry for the reportative evidential -ta2-, which is repeated below.

(46)

[[-ta2-]] = %p. %c: the speaker has reportative evidence relative to p: %p.p

Unlike modals (and epistemic modals, narrowly), evidentials (in Korean, at least) are treated in this thesis as functions from propositions to characters. This difference can account for many syntactic and semantic differences between evidentials and epistemic modals in Korean. For example, evidentials do not appear in embedded contexts, whereas the epistemic modal -keyssdoes, as in (47a) and (47b).

(47) a. *John-un Mary-ka John-Top Mary-Nom

cikum Bill-ul now

ttayli-koiss-ta-nta-ko

Bill-Acc beat-Prog-ta-Decl-Comp

sayngkakha-nta. think-Decl

‘John thinks Mary is beating Bill now’ (with the reportative evidential presupposition) b. John-un

Mary-ka

John-Top Mary-Nom

cikum Bill-ul now

ttayli-koiss-keyss-ta-ko

Bill-Acc beat-Prog-Mod-Decl-Comp

sayngkakha-nta. think-Decl

‘John thinks Mary may be beating Bill now’

! 218

If we assume that attitude predicates such as think take propositions, this difference can be accounted for as follows: -ta2- takes propositions and returns characters and, therefore, attitude predicates cannot apply to these characters (due to type mismatch), whereas -keyss- takes propositions and returns extensions to which attitude predicates can apply via intensional functional application (Heim and Kratzer 1998).90 The difference in the lexical entries of (45) and (46) also explains why the perspective shift is optional with Korean epistemic modals, whereas with Korean evidentials, it is obligatory. To see the optionality of the perspective shift with the Korean modal -keyss-, consider the following question:

(48)

John-i

cikum

cip-ey

iss-keyss-ni?

John-Nom

now

home-Loc

be-Mod-Q?

(lit.) ‘May John be at home now?’ a. Scenario 1: Mary does not know where John is now. She guesses that John may be at home now, so she asks John’s roommate, who she thinks knows where John is now…

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 90

One might think that this account is still problematic, since we still find the declarative -(n)ta- in the embedded clause. The absence of the declarative -ta- in the embedded position results in unacceptable sentences. (47) a'. *John-un Mary-ka John-Top Mary-Nom

cikum now

Bill-ul ttayli-koiss-keyss-+-ko Bill-Acc beat-Prog-Mod-+-Comp

sayngkakha-nta. think-Decl

In the previous section, I analyzed declarative endings as a function from characters to intensions with additional expressive presuppositions. Therefore, it is natural to think that the embedded clause is also a character rather than a proposition. However, the declarative -(n)ta- in the embedded clause is different from other declarative endings in that it does not encode information about the register and relative social status of the conversational participants. This can be shown by replacing the embedded declarative ending -(n)ta- with another declarative ending used in a different register or with conversational participants with different relative social status, which also results in unacceptable sentences. (47) a''. *John-un Mary-ka John-Top Mary-Nom

cikum now

Bill-ul ttayli-koiss-keyss-supnita-ko sayngkakha-nta. Bill-Acc beat-Prog-Mod-Decl-Comp think-Decl

I do not provide a lexical entry for the declarative -(n)ta- in the embedded clause (perhaps it is a simple identity function from propositions to propositions), but I assume that, even though the declarative ending -(n)ta- should be used in the embedded clause, the embedded declarative clause is still a proposition rather than a character.

! 219

b. Scenario 2: Yesterday, Mary saw from John’s schedule that he would be out of town today. Today, Bill is looking for John, saying that he may be at home. Telling him what she saw yesterday, Mary says…

The modal -keyss- can be anchored either to the speaker or to the addressee, depending on the scenarios. Under (48a), where the speaker, Mary, does not have any knowledge about where John is now, the modal -keyss- can be evaluated only against the addressee’s knowledge. However, under (48b), where the speaker, Mary, already has knowledge about where John is now, the modal -keyss- can still be anchored to the speaker.91 This kind of optional shift can be accounted for in terms of Kratzer’s semantics of modals. Two conversational backgrounds, the modal base and the ordering source, are determined in terms of pragmatics and, therefore, may be anchored either to the speaker’s context or to the addressee’s.92 If the proposal made in this thesis is correct, the fundamental difference between evidentials and epistemic modals can be formulated in terms of whether their context variable is bound or remains free. The context variable of evidentials is bound by an operator, whereas the modal base and the ordering source of epistemic modals can be freely anchored either to the speaker’s context or to the addressee’s.

5.4.2. Cross-linguistic variations of evidentiality and epistemic modality

Many researchers investigate the nature of evidentiality in its relationship to epistemic modality, mostly because, as I have pointed out, both evidentiality and epistemic modality !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 91

In this case, the question becomes rhetorical. Since the speaker already has knowledge about where John is now (even though this knowledge remains partial since, in the given scenario, Mary only knows that, according to John’s schedule, he would not be at home right know and does not know exactly where John is or whether John has changed his schedule), it is unnatural for a speaker to ask a question about what the speaker himself/herself already knows, unless it carries a rhetorical implication such as ‘I think John is not at home now’. 92 See also Papafragou (2006) and Portner (2009) for the indexicality of modals and so-called subjective/objective interpretations of epistemic modals.

! 220

concern the speaker’s state of knowledge. This is the main reason why in many languages, it is difficult to distinguish one category from another. Some researchers claim that the two categories are distinct, but others claim that there is overlap between them and even claim that one category belongs to the other. I am unable to make a fully developed proposal about this issue in this thesis. Therefore, in this subsection, I only discuss how evidentials and epistemic modals behave differently in questions and make some speculations about how to account for cross-linguistic variation in the perspective shift of evidentials and epistemic modals in questions. One might simply think that in languages that do not exhibit narrow evidentiality, epistemic modals can either be anchored to the speaker or to the addressee in questions, just like the Korean epistemic modal -keyss-. However, this conclusion is a bit hasty if we consider the distinction between the subjective interpretation and the objective interpretation of epistemic modality (e.g., Lyons 1977, Papafragou 2006). Consider (49).

(49)

It may rain tomorrow. (Lyons 1977; Papafragou 2006: 1691, ex. 14)

According to Lyons (1977), under the subjective reading of may, (50) means that, to the best of the speaker’s knowledge, it is possible that it will rain tomorrow. In contrast, under the objective reading of may, (50) means based on the objective knowledge (i.e., not only the speaker’s knowledge but that of other people and from other reliable source), that it is possible that it will rain tomorrow. Lyons (1977) further proposes that epistemic modals can be embedded or used in questions when they have objective interpretations. For example, in (50), the epistemic modal may reflects the speaker’s view of the possibility of rain tomorrow based on objective knowledge,

! 221

and similarly, in (51), the speaker is asking whether it is possible that it will rain tomorrow based on objective knowledge.

(50) a. It is surprising that it may rain tomorrow, since there was no sign of a cloud all day today. b. The weather forecast told viewers that it may rain tomorrow. (Papafragou 2006: 1692, ex. 17)

(51)

May it rain tomorrow? (Papafragou 2006: 1692, ex. 19)

One might speculate that this distinction is related to whether epistemic modals express indirect evidentiality or epistemic modality. The object interpretation is related to the evidential use of epistemic modals, and the subject interpretation is related to the pure epistemic reading of epistemic modals. As a potential argument against this speculation, von Fintel and Gillies (2007) claim that epistemic modals (at least in English) contain components of indirect evidentiality. With this claim, their proposal seems to amount to further assuming that that every epistemic modal is an (indirect) evidential. However, in this case, it is unclear how to distinguish evidentials from epistemic modals in languages that have both, such as Korean. Further investigation about the judgment of the data is required, and at the moment, I do not have any analysis of the subjective/objective distinction in epistemic modality. Finally, this kind of variation in the interpretation of epistemic modals, and the obligatory versus optional shift of evidentials and epistemic modals in questions may be related to the presence or absence of overt declarative and question endings. Korean is one of few languages (another being Japanese) that have overt declarative endings. As we saw in the previous section, these endings specify in which context the sentence can be uttered felicitously and, in doing so, ! 222

derive intensions from characters. However, English and other Indo-European languages do not have such overt endings, and Cuzco Quechua, where the evidential perspective optionally shifts in questions, does not have such endings, either. It is possible that this difference is also related to the distinction between languages that show the obligatory evidential perspective shift in questions and those that do not. At the moment, I do not have any answer for these questions.

5.5. Reportative evidentiality in Korean noun complements

In this last section, I discuss Kim S.-S.’s (2010) interesting observation about Korean noun complements. Kim points out that there are two types of Korean noun complements with respect to their endings: one with clause/tense markers and one without (hereafter the ‘complex form’ and ‘simple form,’ respectively, following Kim). Kim further observes that the simple form is used “to present a fact or event, but not a proposition,” whereas the complex form is used “to present a proposition or indirect evidence of a fact” (Kim 2010: 11). The example is given below.

(52) a. [sayngsen-i fish-NOM

tha-nun]

naymsay

burn-PNE93

smell

‘the smell of fish burning’ b. *[sayngsen-i fish-NOM

tha-n-ta-nun

naymsay

burn-PRS-DEC-PNE

smell (Kim S.-S. 2010: 2, ex. 3)

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 93

PNE

= prenominal ending!

! 223

(53) a. [Mina-ka Mina-NOM

posek-ul

hwumchi-ess-ta-nun]

somwun/sosik/cwucang

jewelry-ACC

steal-PST-DEC-PNE

rumor/news/claim

‘The rumor/news/claim that Mina stole the jewelry’ b. *[Mina-ka Mina-NOM

posek-ul

hwumchi-n]

somwun/sosik/cwucang

jewelry-ACC

steal-PNE

rumor/news/claim (Kim S.-S. 2010: 2, ex. 4)

(54) a. [cikwu-ka earth-NOM

twunggu-n]

sasil

be.round-PNE

fact

‘the fact that the earth is round’ b. [cikwu-ka earth-NOM

twunggul-ta-nun]

sasil

be.round-DEC-PNE

fact

‘that the earth is round’ (as a proposition) (Kim S.-S. 2010: 2, ex. 5)

In (52), the smell of fish burning is what the speaker directly perceives, and therefore, this smell is ‘presented’ to the speaker as a fact, rather than a proposition. In this case, the complex form in (52b) cannot be used. Rather, only the simple form in (52a) can be used in the prenominal complement. In contrast to (52), in (53), the noun head somwun/sosik/cwucang ‘rumor/news/claim’ indicates that the speaker has heard from someone else that Mina stole the jewelry. Since the speaker has only reportative evidence about the proposition in the noun complement, only the complex form (and not the simple form) can be used. (54) shows that, when a noun such as sasil ‘fact’ can take the noun complement with either the simple form or the complex form, the complement with the simple form indicates the fact that the speaker observes,

! 224

whereas that with the complex form indicates the proposition that the speaker has heard from someone else. Kim S.-S. (2010) speculates that this difference is due to the presence versus absence of a ForceP in the ending. In the simple ending, no ForceP is present, whereas in the complex ending, a ForceP is present. From my perspective, this difference can be accounted for by analyzing -tain the complex form as the reportative evidential -ta- discussed in this thesis. This hypothesis can also be extended to cases where the direct evidential -te- is used in the noun complement, as in (55):

(55)

kwisin-kwa

3 il-kan

nol-ass-te-n

kyenghem

Ghost-with

3 day-during

play-Past-te-PRN

experience

‘The experience that (I) played with a ghost during three days’

At the moment, I do not know whether my proposal has any implications for the syntactic structure Kim proposes nor whether there is any difference between the two proposals in terms of their empirical predictions.

5.6. Summary and conclusions

In this chapter, I provided an overview of various issues related to Korean evidentiality. First, I discussed how sentential endings interact with evidentials in Korean and claimed that sentential endings introduce the felicity conditions surrounding social aspects of the utterance. I also claimed that Korean sentential endings apply to characters, resulting in intensions relative to the utterance context. Furthermore, I also outlined the intimacy reading of Korean reportative ! 225

evidentials and reviewed possible accounts for it. Then, I pointed out that evidential adverbials (and sentential adverbials, probably) in English and Korean show similarities to evidential markers. In addition to this, I briefly discussed the difference between evidentials and epistemic modals in Korean and cross-linguistic variations of evidentials and epistemic modals. Finally, I discussed Kim S.-S.’s (2010) observation concerning evidentiality in Korean noun complements and claimed that my proposal about Korean evidentiality can also be extended to these cases. Many questions raised in this last chapter, as well as some discussed in the previous chapters, remain unresolved, but I believe that my proposal regarding Korean evidentiality and its implications for the semantics of questions can shed light on further investigations about the semantics and pragmatics of evidentials, epistemic modals, interrogatives, and their interactions.

! 226

References Aikhenvald, Alexandra. (2004). Evidentiality. Oxford: Oxford University Press. Amaral, Patricia, Craige Roberts, and E. Allyn Smith. (2007). Review of The Logic of Conventional Implicatures by Chris Potts. Linguistics and Philosophy 30(6): 707–749. Anderson, Lloyd B. (1986). Evidentials, paths of change and mental maps: Typologically regular asymmetries. In W. Chafe and J. Nichols. Eds. Evidentiality: The Linguistic Coding of Epistemology. 273-312. Norwood: Ablex Publishing Corporation. Baker, Mark. (1985). The Mirror Principle and Morphosyntactic Explanation. Linguistic Inquiry 16: 373-415. Bromberger, Sylvain. (1992). On What We Know We Don’t Know: Explanation, Theory, Linguistics, and How Questions Shape Them. Chicago: Chicago University Press. Blain, Eleanor M. and Rose-Marie Déchaine. (2007). Evidential Types: Evidence from Cree Dialects. International Journal of American Linguistics 73: 257-291. Chae, H.-R. (2006). Yunghap hyensangkwa ‘(-ko) ha-’uy kwanlyenseng: [tay]hyeng phyohyenuy pwunsekul cwungsimulo (Fusion and (-ko) ha-: Focusing on the [tae]-type expressions). Language and Information 10(1): 1-20. (Written in Korean) Choi, Kiyong. (2003). Hankwukewa Hayk Itong: Conkyelhyengul Cwungsimulo (Korean and Head Movement: Focusing on Sentential Endings). Studies in Generative Grammar 13(1): 119-142. (Written in Korean) Chung, Joo-Yoon. (2009). A dynamic semantic account of a family of Korean hearsay evidentials. Handout for ISKOL 2009. Chung, Kyung-Sook. (2005). Space in Tense: The Interaction of Tense, Aspect, Evidentiality, and Speech Act in Korean. Doctoral Dissertation, Simon Fraser University. Chung, Kyung-Sook. (2007). Spatial deictic tense and eidentials in Korean. Natural Language Semantics 15: 187-219. Cinque, Guglielmo. (1999). Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Davis, Christopher, Christopher Potts, and Margaret Speas. (2007). The Pragmatic Values of Evidential Sentences. Proceedings in SALT 17. DeLancey, Scott. (1997). Mirativity: the grammatical marking of unexpected information. Linguistic Typology 1: 33-52. DeLancey, Scott. (1997). The mirative and evidentiality. Journal of Pragmatics 33: 369-382. ! 227

Drubig, Hans. (2001). On the syntactic form of epistemic modality. Manuscript. University of Tübingen. Egan, Andy. (2005). Epistemic modals, relativism, and assertion. Philosophical Studies 133: 1-22. Egan, Andy, John Hawthone, and Brian Weatherson. (2005). Epistemic modals in context. In Gerhard Preyer and Georg Peter. Eds. Contextualism in Philosophy. 131-169. Oxford: Oxford University Press. Ernst, Thomas. (2009). Speaker-oriented adverbs. Natural Language and Linguistic Theory 27: 497-544. Faller, Martina. (2002). Semantics and Pragmatics of Evidentials in Cuzco Quechua. Doctoral Dissertation, Stanford. Faller, Martina. (2004). The Deictic Core of ‘Non-Experienced Past’ in Cuzco Quechua. Journal of Semantics 21: 45-85. Faller, Martina. (2006). Evidentiality and Epistemic Modality at the Semantics/Pragmatics Interface. Talk presented at the Workshop on Philosophy and Linguistics. University of Michigan, Ann Arbor, November 3-5, 2006. von Fintel, Kai. (2005). Pragmatics in Linguistic Theory. Lecture Notes, 2005 LSA Summer Institute. Harvard and MIT. von Fintel, Kai. (2008). What is presupposition accommodation, again? Philosophical Perspectives 22: 137-170. von Fintel, Kai, and Irene Heim. (2007). Intensional Semantics: a lecture note. Manuscript. MIT. von Fintel, Kai, and Anthony S. Gillies. (2007). An Opinionated Guide to Epistemic Modality. In Tamar Szabó Gendler and John Hawthone. Eds. Oxford Studies in Epistemology. 32-62. Oxford: Oxford University Press. von Fintel, Kai, and Anthony S. Gillies. (2008). CIA Leaks. Philosophical Review. 117(1): 77-98. von Fintel, Kai, and Anthony S. Gillies. (2010). Must…stay…strong! To appear in Natural Language Semantics. Available at http://mit.edu/fintel/fintel-gillies-2010-mss.pdf. von Fintel, Kai, and Sabine Iatridou. (2003). Epistemic Containment. Linguistic Inquiry 34(2): 173-198. Fitzpatrick, Justin. (2005). The Whys and How Comes of Presupposition and NPI Licensing in Quesitons. In John Alderete, Chung-Hye Han, and Alexei Kochetov. Eds. Proceedings of WCCFL 24. 138-145. Somerville, MA: Cascadilla Proceedings Project. Garrett, Edward. (2001). Evidentiality and Assertion in Tibetan. Doctoral Dissertation. UCLA.

! 228

Gim. Cha-Gyun. (2000). Wulimaluy Sicey Kwucowa Sang Insik. (The structure of tense and aspect interpretation in Korean). Seoul: Thayhaksa. (Written in Korean). Ginzburg, Jonathan. (1996). Interrogatives: Questions, Facts and Dialogue. In Shalom Lappin. Ed. Handbook of Contemporary Semantics. 385-422. London: Blackwell. Green, Mitchell. (2000). Illocutionary force and semantic content. Linguistics and Philosophy 23: 435-473. Grice, H. Paul. (1975). Logic and Conversation. In Peter Cole and Jerry Morgan. Eds. Syntax and Semantics 3: Speech Acts. 41-58. New York: Academic Press. Groenendijk, Jerome, and Martin Stokhof. (1984). Studies on the Semantics of Questions and the Pragmatics of Answers. Doctoral Dissertation, University of Amsterdam. Guerzoni, Elena. (2003). Why Even Ask? Semantics of Questions and Pragmatics of Answers. Doctoral Dissertaiton, MIT. Guerzoni, Elena. (2010). Class notes for the semantics of interrogatives. USC. Guerzoni, Elena, and Lim, Dongsik. (2007). Even if, factivity, and focus, In E. Puig-Waldmüller. Ed. Proceedings of Sinn und Bedeutung 11. 276-290. Barcelona: Universitat Pompeu Fabra. de Haan, Ferdinand. (1998). The category of evidentiality. Unpublished ms., University of New Mexico. Harris, Jesse, and Christopher Potts. (2010). Perspective-shifting with appositives and expressives. Manuscript. UMass and Stanford University. http://semanticsarchive.net/Archive/jAyNWI4N/ Hamblin, C. L. (1973). Questions in Montague English. Foundations of Language 10: 41-53. Heim, Irene, and Kai von Fintel. (2001). Class notes for ‘Advanced Semantics’. MIT. Heim, Irene, and Angelika Kratzer. (1998). Semantics of Generative Grammar. London: Blackwell. Higginbotham, James. (1996). The Semantics of Questions. In Shalom Lappin. Ed. Handbook of Contemporary Semantics. 362-383. London: Blackwell. Higginbotham, James. (2009a). Evidentials: Some Preliminary Distinctions. In Robert J. Stainton and Christopher Viger. Eds. Compositionality, context and Semantic Values: Essays in Honour of Ernie Lepore. 221-235. Dordrecht: Springer. Higginbotham, James. (2009b). Anaphoric Tense. In Higginbotham, James. (2009). Tense, Aspect, and Indexicality. Oxford: Oxford University Press. Horn, Laurence. (1989). A Natural History of Negation. Chicago: University of Chicago Press. ! 229

Izvorski, Roumyana. (1997). The Present Perfect as an Epistemic Modal. In A. Lawson and E. Cho. Eds. The Proceedings of SALT 7. Kaplan, David. (1989). Demonstratives. Eds. Almog, J., Perry, J., and Weinstein, S. Themes from Kaplan. Oxford: Oxford University Press. Reprinted in ed. Davidson, Matthews. 2007. On Sense and Direct Reference. 720-782. Boston: McGraw Hill. Karttunen, Lauri. (1977). Syntax and semantics in questions. Linguistics and Philosophy 1: 3-44. Kim, Mi-Ran, Shin, Dong-Hyun, Choe, Jae-Woong, and Kim, Kee-Ho. (2000). Chocemkwa Kwanlyentoyn Uymwunmwun Ekyang Phaythen Silhem (Pitch Patterns of Interrogative Sentences in relation to the Focus). Umsengkwahak (Phonetic Science) 7: 217. (Written in Korean) Kim, Nam-Kil. (1990). Hearsay Evidential in Korean. Icwungenehakhoyji (Journal of the Korean Society of Bilingualism) 7: 418-434. Kim, Nam-Kil. (2001). Reportative evidential in Korean. Journal of the International Circle of Korean Linguistics 10: 105-124. Kim, Shin-Sook. (2008). Noun Complements and Clause Types in Korean (and Japanese). Papers presented in the 18th Japanese/Korean Linguistics Conference. City University of New York. Klinedinst, Nathan. (2005). Plurals, possibilities, and conjunctive disjunction, paper presented in Sinn und Bedeutung 10. Ko, Heejeong. (2005). Syntax of why-in-situ: merge into [spec,CP] in the overt syntax. Natural Lauguage and Linguistic Theory 23: 867-916. Ko, Heejeong. (2006). On the Structural Height of Reason Wh-Adverbials: Acquisition and Consequences. In Lisa Lai-Shen Cheng and Norbert Corver. Eds. Wh-Movement: Moving on. 319-349. Cambridge: MIT Press. Ko, Youngkeun. (2007). Hankwukeuy Sicey Sepep Tongcaksang. (Tense, Mood, and Aspect in Korean). Seoul: Taehaksa. (Written in Korean). Kratzer, Angelika. (1977). What ‘must’ and ‘can’ must and can mean. Linguistics and Philosophy 1: 337-355. Kratzer, Angelika. (1981). The notional category of modality. H. Eickmeyer & H. Rieser. Eds. Words, Worlds and Contexts. 38-74. Berlin: Walter de Gruyter. Kratzer, Angelika. (1991). Modality. In Arnim von Stechow and Dieter Wunderlich, Eds, Semantics: An International Handbook of Contemporary Research. 639-650. Berlin: de Gruyter. Kripke, Saul. (1980). Naming and Necessity. Cambridge: Harvard University Press. ! 230

Kwon, Iksoo. (2010). I Guess Korean Has Some Mirative Markers: -Napo- and -Nmoyang Lee, Jungmee. (2010). The evidential meaning of -te in Korean: A compositional analysis. Handout presented in the 84th annual meeting of the Linguistic Society of America Lim, Dongsik. (2009). Korean Non-Final –ta- as an Indirect Evidential Marker. In Proceedings of WECOL 2008, 175-186. Lyons, John. (1977). Semantics, 2 vols. Cambridge: Cambridge University Press. Matthewson, Lisa, Henry Davis, and Hotze Rullman. (2007). Evidentials as epistemic modals: evidence from St’át’imcets. Linguistic Variation Yearbook 7: 201-254. McCready, Eric and Norry Ogata. (2007). Evidentiality, modality and probability. Linguistics and Philosophy 30: 147-206. Murray, Sarah E. (2009a). A Hamblin Semantics for Evidentials. In Proceedings of Semantics and Linguistics Theory 19. Murray, Sarah E. (2009b). Evidentials and Questions in Cheyenne. In Proceedings of SULA 5: Semantics of Under-Represented Languages in the America. Pak, Miok, Paul Portner, and Raffaella Zanuttini. (2005). What Korean Promissives tell us about Jussive Clause Types. Paper presented in Colloque de syntaxe and sémantique à Paris 2005. Pak, Miok, Paul Portner, and Raffaella Zanuttini. (2007). Agreement and the Subjects of Jussive Clauses in Korean. Eds. E. Efner and M. Walkow. Proceedings of North East Linguistics Society 37. Palmer, Frank. (2001). Mood and Modality (2nd edition). Cambridge: Cambridge University Press. Papafragou, Anna. (2006). Epistemic modality and truth conditions. Lingua 116: 1688-1702. Park, Jae-Yon. (1998). Hyentaykwuke Panmalchey Congkyelemi Yenkwu (A Study on Sentence Final Endings of Panmal Style in Contemporary Korean). Master’s Thesis, Seoul National University. (Written in Korean) Partee, Barbara H. (1984). Nominal and Temporal Anaphora. Linguistics and Philosophy 7(3): 243-287. Portner, Paul. (2009). Modality. Oxford: Oxford University Press. Potts, Christopher. (2005). The Logic of Conventional Implicature. Oxford: Oxford University Press. Potts, Christopher. (2007a). The Dimensions of Quotation. In Chris Barker and Pauline Jacobson. Eds. Direct Compositionality. 405-431. Oxford: Oxford University Press. ! 231

Potts, Christopher. (2007b). The expressive dimension. Theoretical Linguistics 33(2): 165-198. Potts, Christopher. (2007c). The centrality of expressive indices. Theoretical Linguistics 33(2): 255-268. Rizzi, Luigi. (1999). On the Position of “Int(errogative)” in the Left Periphery of the Clause. ms., Università di Siena. Roberts, Craige. (1987). Modal Subordination, Anaphora, and Distributivity. Doctoral Dissertation, University of Massachusetts Amherst. Roberts, Craige. (1989). Modal subordination and pronominal anaphora in discourse. Linguistics and Philosophy 12: 683-721. Romero, Maribel and Chung-Hye Han. (2004). On negative yes/no questions. Linguistics and Philosophy 27: 609-658. Rooryck, Johan. (2001a). Evidentiality: Part I. Glot International 5(4): 125-133. Rooryck, Johan. (2001b). Evidentiality: Part II. Glot International 5(5): 161-168. Rooth, Mats. (1985). Association with Focus. Doctoral Dissertation, University of Massachusetts Amherst. Rooth, Mats. (1996). Focus. In Shalom Lappin. Eds. The Handbook of Contemporary Semantic Theory. 271-297. Oxford: Blackwell. Rullmann, Hotze, Lisa Matthewson, and Henry Davis. (2008). Modals as distributive indefinites. Natural Language Semantics. 16: 317-357. Saito, Mamoru. (1994). Additional-wh effects and the adjunction site theory. Journal of East Asian Linguistics 3: 195-240. Sauerland, Uli. (2007). Beyond unpluggability. Theoretical Linguistics 33(2): 231-236. Schlenker, Philippe. (2003). A Plea for Monsters. Linguistics and Philosophy 26: 29-120. Schlenker, Philippe. (2007). The expressive presupposition. Theoretical Linguistics 33(2): 237245. Searle, John. (1969). Speech Acts. Cambridge: Cambridge University Press. Searle, John, and Daniel Vanderveken. (1985). Foundations of Illocutionary Logic. Cambridge: Cambridge University Press. Sells, Peter. (1995). Korean and Japanese morphology from a lexical perspective. Linguistic Inquiry 20: 365-424. ! 232

Simons, Mandy. (2007). Observations on embedding verbs, evidentiality, and presupposition. Lingua 117: 1034-1056. Soames, Scott. (1982). How presuppositions are inherited: a solution to the projection problem. Linguistic Inquiry 13: 483-545. Soames, Scott. (1989). Presupposition. Eds. Gabbay, D., and Guenthner, F. Handbook of Philosophical Logic IV, 553-616. Dordrecht: Reidel Publishing Company. Sohn, Ho-Min. (1994). Korean. London: Routledge. Song, Jae-Mog. (1997). Tense, Aspect and Modality in Khalkha Mongolian. Doctoral Dissertation, School of Oriental and Asian Studies, London. Song, Jae-Mog. (1998). Anmaycumssikkuth ‘-te-’uy uymi kinungey tayhaye: yuhyengloncek kwancemeyse. (On the semantic function of the pre-final ending –te-: from a typological perspective). Kwukehak (Korean Linguistics) 32: 135-169. (Written in Korean) Stalnaker, Robert. (1974). Pragmatic Presuppositions. In M. Munitz and P. Unger. Eds. Semantics and Philosophy. New York: New York University Press. Reprinted in Steven Davis. Eds. (1991). Pragmatics: A Reader. Oxford: Oxford University Press. Stalnaker, Robert. (2002). Common ground. Linguistics and Philosophy 25: 701-721. Tenny, Carol. (2006). Evidentiality, experiencers, and the syntax of sentience in Japanese. Journal of East Asian Linguistics 15: 245-288. Yalcin, Seth. (2007). Epistemic Modals. Mind 116: 983-1026. Yoon, Jeong-Me. (1990). Verb movement and the structure of IP in Korean. Language Research 26: 343-371. Yu-Cho, Young-Mee, and Peter Sells. (1995). A lexical account of inflectional suffixes in Korean. Journal of East Asian Linguistics 4: 119-174.

! 233